{
 "meta": {
  "as_of": "2026-10-13",
  "schema": "Every figure is {value, source, as_of, note?}. value null = not verified; the site prints an em dash. source is a key into 'sources'.",
  "benchmarks": {
   "aa_index": "Artificial Analysis Intelligence Index v4.3.2 — independent. Weighted mix: agents 30% (AA-Briefcase, GDPval-AA, AutomationBench-AA), coding 20% (Terminal-Bench 4.0, SciCode), general 30% (AA-Omniscience, GDP.pdf, AA-LCR), scientific reasoning 20% (HLE, CritPt).",
   "aa_cost_per_task": "Artificial Analysis — average USD cost to run one Intelligence Index task at the listed reasoning setting. Independent.",
   "speed_tps": "Artificial Analysis — median output tokens per second on the vendor's own API. Independent.",
   "ttft_s": "Artificial Analysis — seconds until the first answer token, including thinking time. Independent.",
   "terminal_bench_4": "Terminal-Bench 4.0 resolution rate (agentic coding in a terminal). Figures here are vendor-reported.",
   "gdpval_aa_elo": "GDPval-AA v2.1 Elo (real-world knowledge-work tasks), run by Artificial Analysis; figures here were read from vendor announcement pages.",
   "deepswe_v1_1": "DeepSWE v1.1 (software engineering). Vendor-reported.",
   "swe_bench_verified": "SWE-bench Verified. Not populated — see field notes.",
   "arena_elo": "Arena (formerly LMArena) text Elo. Not populated — see field notes."
  }
 },
 "sources": {
  "aa_lb": {
   "name": "Artificial Analysis — LLM leaderboard (Intelligence Index v4.3.2)",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_method": {
   "name": "Artificial Analysis — Intelligence Index methodology",
   "url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_sts": {
   "name": "Artificial Analysis — Speech-to-Speech leaderboard",
   "url": "https://artificialanalysis.ai/speech-to-speech",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_opus55": {
   "name": "Artificial Analysis — Claude Opus 5.5",
   "url": "https://artificialanalysis.ai/models/claude-opus-5-5",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_haiku55": {
   "name": "Artificial Analysis — Claude Haiku 5.5",
   "url": "https://artificialanalysis.ai/models/claude-haiku-5-5",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_luna": {
   "name": "Artificial Analysis — GPT-6 Luna",
   "url": "https://artificialanalysis.ai/models/gpt-6-luna",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_argon": {
   "name": "Artificial Analysis — Gemini 4 Argon",
   "url": "https://artificialanalysis.ai/models/gemini-4-argon",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_flash38": {
   "name": "Artificial Analysis — Gemini 3.8 Flash",
   "url": "https://artificialanalysis.ai/models/gemini-3-8-flash",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_gemma": {
   "name": "Artificial Analysis — Gemma 4 31B",
   "url": "https://artificialanalysis.ai/models/gemma-4-31b",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_wonder": {
   "name": "Artificial Analysis — Muse Wonder",
   "url": "https://artificialanalysis.ai/models/muse-wonder",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_spark": {
   "name": "Artificial Analysis — Muse Spark 1.3",
   "url": "https://artificialanalysis.ai/models/muse-spark-1-3",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_glimmer": {
   "name": "Artificial Analysis — Muse Glimmer",
   "url": "https://artificialanalysis.ai/models/muse-glimmer",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_grok": {
   "name": "Artificial Analysis — Grok 4.7",
   "url": "https://artificialanalysis.ai/models/grok-4-7",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_ml4": {
   "name": "Artificial Analysis — Mistral Large 4",
   "url": "https://artificialanalysis.ai/models/mistral-large-4",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_dsflash": {
   "name": "Artificial Analysis — DeepSeek V4.1 Flash",
   "url": "https://artificialanalysis.ai/models/deepseek-v4-1-flash",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_k3": {
   "name": "Artificial Analysis — Kimi K3",
   "url": "https://artificialanalysis.ai/models/kimi-k3",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_qmax": {
   "name": "Artificial Analysis — Qwen3.8 Max",
   "url": "https://artificialanalysis.ai/models/qwen3-8-max",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_q27": {
   "name": "Artificial Analysis — Qwen3.8 27B",
   "url": "https://artificialanalysis.ai/models/qwen3-8-27b",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_nova": {
   "name": "Artificial Analysis — Nova 2.0 Pro",
   "url": "https://artificialanalysis.ai/models/nova-2-0-pro",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_glm": {
   "name": "Artificial Analysis — GLM-5.3",
   "url": "https://artificialanalysis.ai/models/glm-5-3",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_mimo": {
   "name": "Artificial Analysis — MiMo-V2.6-Pro",
   "url": "https://artificialanalysis.ai/models/mimo-v2-6-pro",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "aa_mercury": {
   "name": "Artificial Analysis — Mercury 2.5",
   "url": "https://artificialanalysis.ai/models/mercury-2-5",
   "kind": "independent",
   "fetched": "2026-10-13"
  },
  "anth_models": {
   "name": "Anthropic — Models overview (pricing, context, IDs)",
   "url": "https://platform.claude.com/docs/en/about-claude/models/overview",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "anth_opus": {
   "name": "Anthropic — Claude Opus 5.5 announcement",
   "url": "https://www.anthropic.com/claude-opus-5-5",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "anth_sonnet": {
   "name": "Anthropic — Claude Sonnet 5.5 announcement",
   "url": "https://www.anthropic.com/claude-sonnet-5-5",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "anth_haiku": {
   "name": "Anthropic — Claude Haiku 5.5 announcement",
   "url": "https://www.anthropic.com/claude-haiku-5-5",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "anth_fable": {
   "name": "Anthropic — Claude Fable 5.1 and Mythos 5.1 announcement",
   "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "oai_pricing": {
   "name": "OpenAI — API pricing",
   "url": "https://developers.openai.com/api/docs/pricing",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "oai_models": {
   "name": "OpenAI — Models",
   "url": "https://developers.openai.com/api/docs/models",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "oai_astra": {
   "name": "OpenAI — GPT-6 Astra model page",
   "url": "https://developers.openai.com/api/docs/models/gpt-6-astra",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "oai_sol": {
   "name": "OpenAI — GPT-6.1 Sol model page",
   "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "goog_pricing": {
   "name": "Google — Gemini API pricing",
   "url": "https://ai.google.dev/gemini-api/docs/pricing",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "goog_argon": {
   "name": "Google — Gemini 4 Argon announcement",
   "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "goog_flash": {
   "name": "Google — Gemini 3.8 Flash announcement",
   "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "xai_models": {
   "name": "xAI — Models and pricing",
   "url": "https://docs.x.ai/docs/models",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "xai_grok47": {
   "name": "xAI — Grok 4.7 announcement",
   "url": "https://x.ai/news/grok-4-7",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "meta_wonder": {
   "name": "Meta — Muse Wonder announcement",
   "url": "https://research.meta.ai/blog/muse-wonder",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "meta_dev": {
   "name": "Meta — Muse model API page",
   "url": "https://dev.meta.ai/models/muse-spark/",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "mistral_l4": {
   "name": "Mistral — Mistral Large 4 announcement",
   "url": "https://mistral.ai/news/mistral-large-4/",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "ds_pricing": {
   "name": "DeepSeek — API pricing",
   "url": "https://api-docs.deepseek.com/quick_start/pricing",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "ds_news": {
   "name": "DeepSeek — V4.1 Flash release note",
   "url": "https://api-docs.deepseek.com/news/news260910",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "kimi_pricing": {
   "name": "Moonshot / Kimi — API pricing",
   "url": "https://platform.kimi.ai/docs/pricing/chat",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "ali_pricing": {
   "name": "Alibaba Cloud — Model Studio pricing",
   "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "ms_mai": {
   "name": "Microsoft AI — streaming transcription and MAI-Voice-2.1 announcement",
   "url": "https://microsoft.ai/news/our-first-streaming-transcription-model/",
   "kind": "primary",
   "fetched": "2026-10-13"
  },
  "benchlm": {
   "name": "BenchLM — model release tracker (aggregator; dates are claims)",
   "url": "https://benchlm.ai/md/model-updates.md",
   "kind": "aggregator",
   "fetched": "2026-10-13"
  },
  "euronews": {
   "name": "Euronews — AI labs sued over alleged slowdown pact (21 Sep 2026)",
   "url": "https://www.euronews.com/next/2026/09/21/for-once-big-tech-agreed-on-something-now-its-being-sued-for-it",
   "kind": "press",
   "fetched": "2026-10-13"
  }
 },
 "models": [
  {
   "id": "claude-opus-5-5",
   "name": "Claude Opus 5.5",
   "company": "anthropic",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "aa_opus55",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-22",
    "source": "anth_opus",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 4.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 20.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 58,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 5.98,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 95.2,
    "source": "aa_opus55",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 691.73,
    "source": "aa_opus55",
    "as_of": "2026-10-13",
    "note": "At max effort the model thinks for minutes before the first token; see variants for lower effort."
   },
   "terminal_bench_4": {
    "value": 66.4,
    "source": "anth_opus",
    "as_of": "2026-10-13",
    "note": "Vendor-reported, xhigh effort."
   },
   "gdpval_aa_elo": {
    "value": 1846,
    "source": "anth_opus",
    "as_of": "2026-10-13",
    "note": "GDPval-AA v2.1, run by Artificial Analysis, published on Anthropic's page."
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 56,
     "aa_cost_per_task": 3.46,
     "speed_tps": 81,
     "ttft_s": 159.41,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "high",
     "aa_index": 54,
     "aa_cost_per_task": 1.82,
     "speed_tps": 82,
     "ttft_s": 52.59,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "medium",
     "aa_index": 51,
     "aa_cost_per_task": 1.34,
     "speed_tps": 79,
     "ttft_s": 25.47,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "claude-fable-5-1",
   "name": "Claude Fable 5.1",
   "company": "anthropic",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-01",
    "source": "benchlm",
    "as_of": "2026-10-13",
    "note": "Anthropic's page says 'September 2026'; its retirement commitment ('not sooner than September 1, 2027') matches the aggregator date."
   },
   "context_tokens": {
    "value": 1000000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 10.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 50.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 53,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 7.63,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 68,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 246.61,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": 55.8,
    "source": "anth_fable",
    "as_of": "2026-10-13",
    "note": "Vendor-reported. xAI's Grok 4.7 page lists 57.9% for Fable 5.1 — sources disagree."
   },
   "gdpval_aa_elo": {
    "value": 1735,
    "source": "anth_opus",
    "as_of": "2026-10-13",
    "note": "GDPval-AA v2.1. Anthropic's Fable page shows 1853 on the older v2 scale."
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 53,
     "aa_cost_per_task": 5.98,
     "speed_tps": 62,
     "ttft_s": 113.74,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "high",
     "aa_index": 51,
     "aa_cost_per_task": 3.91,
     "speed_tps": 52,
     "ttft_s": 17.98,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "medium",
     "aa_index": 49,
     "aa_cost_per_task": 2.98,
     "speed_tps": 54,
     "ttft_s": 10.27,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": [
    "Sibling model Claude Mythos 5.1 is restricted to verified cyber and life-science programs and is not listed here."
   ]
  },
  {
   "id": "claude-sonnet-5-5",
   "name": "Claude Sonnet 5.5",
   "company": "anthropic",
   "tier": "mid",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-28",
    "source": "anth_sonnet",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 2.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 10.0,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 56,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 5.46,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 135,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 451.91,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": 70.6,
    "source": "anth_sonnet",
    "as_of": "2026-10-13",
    "note": "Vendor-reported."
   },
   "gdpval_aa_elo": {
    "value": 1844,
    "source": "anth_sonnet",
    "as_of": "2026-10-13",
    "note": "Anthropic's Haiku 5.5 page lists 1840 for the same model — sources disagree by 4 points."
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 52,
     "aa_cost_per_task": 2.01,
     "speed_tps": 103,
     "ttft_s": 36.14,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "claude-haiku-5-5",
   "name": "Claude Haiku 5.5",
   "company": "anthropic",
   "tier": "small",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "aa_haiku55",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-10-07",
    "source": "anth_haiku",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "anth_models",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 0.1,
    "source": "anth_haiku",
    "as_of": "2026-10-13",
    "note": "Prompts up to 100k tokens. Above that: $0.50."
   },
   "price_out": {
    "value": 0.5,
    "source": "anth_haiku",
    "as_of": "2026-10-13",
    "note": "Prompts up to 100k tokens. Above that: $2.50."
   },
   "aa_index": {
    "value": 43,
    "source": "aa_haiku55",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.21,
    "source": "aa_haiku55",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 237.9,
    "source": "aa_haiku55",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 309.8,
    "source": "aa_haiku55",
    "as_of": "2026-10-13",
    "note": "Max effort; Artificial Analysis shows this far above the class median."
   },
   "terminal_bench_4": {
    "value": 39.2,
    "source": "anth_haiku",
    "as_of": "2026-10-13",
    "note": "Vendor-reported."
   },
   "gdpval_aa_elo": {
    "value": 1620,
    "source": "anth_haiku",
    "as_of": "2026-10-13"
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "gpt-6-astra",
   "name": "GPT-6 Astra",
   "company": "openai",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "oai_astra",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "oai_astra",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-03",
    "source": "benchlm",
    "as_of": "2026-10-13",
    "note": "Aggregator date; OpenAI's model page carries no release date."
   },
   "context_tokens": {
    "value": 1050000,
    "source": "oai_astra",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "oai_astra",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 10.0,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 50.0,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 53,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 3.26,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 63,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 326.13,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": 57.9,
    "source": "anth_opus",
    "as_of": "2026-10-13",
    "note": "As reported by OpenAI, republished on Anthropic's Opus 5.5 page; OpenAI's own post returned HTTP 403 to our fetch."
   },
   "gdpval_aa_elo": {
    "value": 1542,
    "source": "anth_opus",
    "as_of": "2026-10-13",
    "note": "GDPval-AA v2.1; the same figure appears on xAI's Grok 4.7 page."
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 52,
     "aa_cost_per_task": 2.31,
     "speed_tps": 59,
     "ttft_s": 160.99,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "high",
     "aa_index": 51,
     "aa_cost_per_task": 1.73,
     "speed_tps": 55,
     "ttft_s": 94.34,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "medium",
     "aa_index": 50,
     "aa_cost_per_task": 1.54,
     "speed_tps": 55,
     "ttft_s": 5.94,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "gpt-6-1-sol",
   "name": "GPT-6.1 Sol",
   "company": "openai",
   "tier": "mid",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "oai_sol",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "oai_sol",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-29",
    "source": "benchlm",
    "as_of": "2026-10-13",
    "note": "Aggregator date; OpenAI's announcement returned HTTP 403 to our fetch."
   },
   "context_tokens": {
    "value": 1050000,
    "source": "oai_sol",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "oai_sol",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 2.0,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 10.0,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 52,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.72,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 65,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 321.8,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 51,
     "aa_cost_per_task": 0.39,
     "speed_tps": 59,
     "ttft_s": 173.89,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "high",
     "aa_index": 50,
     "aa_cost_per_task": 0.32,
     "speed_tps": 57,
     "ttft_s": 55.38,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "gpt-6-luna",
   "name": "GPT-6 Luna",
   "company": "openai",
   "tier": "small",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "oai_models",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-22",
    "source": "aa_luna",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1050000,
    "source": "oai_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 128000,
    "source": "oai_models",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 0.1,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 0.5,
    "source": "oai_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 38,
    "source": "aa_luna",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "speed_tps": {
    "value": 134.2,
    "source": "aa_luna",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 113.29,
    "source": "aa_luna",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": 16.4,
    "source": "anth_haiku",
    "as_of": "2026-10-13",
    "note": "Published on Anthropic's Haiku 5.5 comparison table."
   },
   "gdpval_aa_elo": {
    "value": 1437,
    "source": "anth_haiku",
    "as_of": "2026-10-13"
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "gemini-4-argon",
   "name": "Gemini 4 Argon",
   "company": "google",
   "tier": "flagship",
   "aa_setting": "high",
   "open_weights": {
    "value": false,
    "source": "aa_argon",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image",
     "video"
    ],
    "source": "goog_argon",
    "as_of": "2026-10-13",
    "note": "Google's post describes chart and long-video understanding; it does not print a modality list."
   },
   "released": {
    "value": "2026-09-30",
    "source": "goog_argon",
    "as_of": "2026-10-13",
    "note": "Announcement date. Google says it is 'rolling out soon' — not generally available; not yet on the Gemini API pricing page."
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_argon",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 2.0,
    "source": "goog_argon",
    "as_of": "2026-10-13",
    "note": "Introductory. Google states $4.00 after the introductory period."
   },
   "price_out": {
    "value": 10.0,
    "source": "goog_argon",
    "as_of": "2026-10-13",
    "note": "Introductory. Google states $20.00 after the introductory period."
   },
   "aa_index": {
    "value": 53,
    "source": "aa_argon",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 1.99,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "No public API speed measured yet."
   },
   "ttft_s": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "No public API latency measured yet."
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": 77.9,
    "source": "goog_argon",
    "as_of": "2026-10-13",
    "note": "Vendor-reported; Google calls it state of the art."
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "gemini-3-8-flash",
   "name": "Gemini 3.8 Flash",
   "company": "google",
   "tier": "mid",
   "aa_setting": "high",
   "open_weights": {
    "value": false,
    "source": "aa_flash38",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-02",
    "source": "goog_flash",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_flash38",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.75,
    "source": "goog_pricing",
    "as_of": "2026-10-13",
    "note": "Introductory through 31 Dec 2026; $1.50 from 1 Jan 2027."
   },
   "price_out": {
    "value": 3.75,
    "source": "goog_pricing",
    "as_of": "2026-10-13",
    "note": "Introductory through 31 Dec 2026; $7.50 from 1 Jan 2027."
   },
   "aa_index": {
    "value": 41,
    "source": "aa_flash38",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 1.24,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 141.4,
    "source": "aa_flash38",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 22.67,
    "source": "aa_flash38",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "gemini-3-5-flash-lite",
   "name": "Gemini 3.5 Flash-Lite",
   "company": "google",
   "tier": "small",
   "aa_setting": "default",
   "open_weights": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.3,
    "source": "goog_pricing",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 2.5,
    "source": "goog_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 22,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.19,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 408,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 7.92,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "gemma-4-31b",
   "name": "Gemma 4 31B",
   "company": "google",
   "tier": "open",
   "aa_setting": "reasoning",
   "open_weights": {
    "value": true,
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "Apache 2.0",
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-04-02",
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 256000,
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "Artificial Analysis displays $0.00, which reads as a placeholder; not used."
   },
   "price_out": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "Artificial Analysis displays $0.00, which reads as a placeholder; not used."
   },
   "aa_index": {
    "value": 15,
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "Leaderboard shows $0.00; not used."
   },
   "speed_tps": {
    "value": 35.5,
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 1.03,
    "source": "aa_gemma",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "muse-wonder-full",
   "name": "Muse Wonder Full",
   "company": "meta",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "meta_wonder",
    "as_of": "2026-10-13",
    "note": "Available only to Muse Code subscribers; Meta states no open weights."
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-10-12",
    "source": "meta_wonder",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "No API price published."
   },
   "price_out": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "No API price published."
   },
   "aa_index": {
    "value": 56,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "speed_tps": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "ttft_s": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "muse-wonder",
   "name": "Muse Wonder",
   "company": "meta",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "aa_wonder",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-10-12",
    "source": "meta_wonder",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_wonder",
    "as_of": "2026-10-13",
    "note": "The Artificial Analysis leaderboard row shows 1.05M — sources disagree."
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 4.0,
    "source": "aa_wonder",
    "as_of": "2026-10-13",
    "note": "Meta's announcement prints no price; figure is from Artificial Analysis."
   },
   "price_out": {
    "value": 20.0,
    "source": "aa_wonder",
    "as_of": "2026-10-13",
    "note": "Meta's announcement prints no price; figure is from Artificial Analysis."
   },
   "aa_index": {
    "value": 52,
    "source": "aa_wonder",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 4.52,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 101.4,
    "source": "aa_wonder",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 19.73,
    "source": "aa_wonder",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "muse-spark-1-3",
   "name": "Muse Spark 1.3",
   "company": "meta",
   "tier": "mid",
   "aa_setting": "max",
   "open_weights": {
    "value": false,
    "source": "aa_spark",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image",
     "video"
    ],
    "source": "meta_dev",
    "as_of": "2026-10-13"
   },
   "released": {
    "value": "2026-09-02",
    "source": "aa_spark",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "meta_dev",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 1.25,
    "source": "meta_dev",
    "as_of": "2026-10-13",
    "note": "Standard tier. A 'contributor' tier (data used to improve Meta products) is $0.10."
   },
   "price_out": {
    "value": 4.25,
    "source": "meta_dev",
    "as_of": "2026-10-13",
    "note": "Standard tier. Contributor tier is $0.20."
   },
   "aa_index": {
    "value": 48,
    "source": "aa_spark",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 1.6,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 123.9,
    "source": "aa_spark",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 40.76,
    "source": "aa_spark",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "xhigh",
     "aa_index": 45,
     "aa_cost_per_task": 1.37,
     "speed_tps": 245,
     "ttft_s": 22.34,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "muse-glimmer",
   "name": "Muse Glimmer 30B",
   "company": "meta",
   "tier": "open",
   "aa_setting": "high",
   "open_weights": {
    "value": true,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "Apache 2.0",
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-08-10",
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 131000,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.325,
    "source": "aa_glimmer",
    "as_of": "2026-10-13",
    "note": "Hosted-API price tracked by Artificial Analysis."
   },
   "price_out": {
    "value": 1.35,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 17,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.06,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 133.8,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 0.99,
    "source": "aa_glimmer",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "grok-4-7",
   "name": "Grok 4.7",
   "company": "xai",
   "tier": "flagship",
   "aa_setting": "xhigh",
   "open_weights": {
    "value": false,
    "source": "aa_grok",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text"
    ],
    "source": "xai_models",
    "as_of": "2026-10-13",
    "note": "xAI's docs mention image input without naming which models."
   },
   "released": {
    "value": "2026-09-21",
    "source": "xai_grok47",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 500000,
    "source": "xai_models",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 2.0,
    "source": "xai_models",
    "as_of": "2026-10-13",
    "note": "Prompts under 200k tokens. $4.00 at 200k and above."
   },
   "price_out": {
    "value": 6.0,
    "source": "xai_models",
    "as_of": "2026-10-13",
    "note": "Prompts under 200k tokens. $12.00 at 200k and above."
   },
   "aa_index": {
    "value": 46,
    "source": "aa_grok",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 3.74,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 83.8,
    "source": "aa_grok",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 86.95,
    "source": "aa_grok",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": 37.6,
    "source": "xai_grok47",
    "as_of": "2026-10-13",
    "note": "Vendor-reported."
   },
   "gdpval_aa_elo": {
    "value": 1695,
    "source": "xai_grok47",
    "as_of": "2026-10-13",
    "note": "Read from a chart on xAI's page; label mapping inferred from layout."
   },
   "deepswe_v1_1": {
    "value": 71.0,
    "source": "xai_grok47",
    "as_of": "2026-10-13",
    "note": "Vendor-reported, high effort."
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [
    {
     "setting": "high",
     "aa_index": 46,
     "aa_cost_per_task": 2.73,
     "speed_tps": 74,
     "ttft_s": 46.44,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    },
    {
     "setting": "low",
     "aa_index": 42,
     "aa_cost_per_task": 1.25,
     "speed_tps": 65,
     "ttft_s": 6.97,
     "source": "aa_lb",
     "as_of": "2026-10-13"
    }
   ],
   "notes": []
  },
  {
   "id": "mistral-large-4",
   "name": "Mistral Large 4",
   "company": "mistral",
   "tier": "flagship",
   "aa_setting": "preview",
   "open_weights": {
    "value": false,
    "source": "aa_ml4",
    "as_of": "2026-10-13",
    "note": "Public preview. Mistral says weights will be released by the end of October 2026; not out as of this date."
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": [
     "text",
     "image"
    ],
    "source": "mistral_l4",
    "as_of": "2026-10-13",
    "note": "Mistral says 'natively multimodal' input without listing types."
   },
   "released": {
    "value": "2026-10-06",
    "source": "mistral_l4",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 524000,
    "source": "aa_ml4",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 1.36,
    "source": "mistral_l4",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 4.18,
    "source": "mistral_l4",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 38,
    "source": "aa_ml4",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 1.13,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "ttft_s": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "terminal_bench_4": {
    "value": 28.3,
    "source": "mistral_l4",
    "as_of": "2026-10-13",
    "note": "Vendor-reported."
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": 61.7,
    "source": "mistral_l4",
    "as_of": "2026-10-13",
    "note": "Vendor-reported."
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "mistral-small-4",
   "name": "Mistral Small 4",
   "company": "mistral",
   "tier": "small",
   "aa_setting": "reasoning",
   "open_weights": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "context_tokens": {
    "value": 256000,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_out": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "aa_index": {
    "value": 11,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.02,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 158,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 0.77,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "deepseek-v4-1-flash",
   "name": "DeepSeek V4.1 Flash",
   "company": "deepseek",
   "tier": "mid",
   "aa_setting": "max",
   "open_weights": {
    "value": true,
    "source": "aa_dsflash",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "MIT",
    "source": "aa_dsflash",
    "as_of": "2026-10-13",
    "note": "DeepSeek's release note links the weights but does not name the license."
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-10",
    "source": "ds_news",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1000000,
    "source": "ds_pricing",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 384000,
    "source": "ds_pricing",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 0.3,
    "source": "ds_pricing",
    "as_of": "2026-10-13",
    "note": "Peak rate, cache miss. Off-peak is $0.15."
   },
   "price_out": {
    "value": 1.2,
    "source": "ds_pricing",
    "as_of": "2026-10-13",
    "note": "Peak rate. Off-peak is $0.60."
   },
   "aa_index": {
    "value": 39,
    "source": "aa_dsflash",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.27,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 197.3,
    "source": "aa_dsflash",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 1.06,
    "source": "aa_dsflash",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "deepseek-v4-pro-0813",
   "name": "DeepSeek V4 Pro 0813",
   "company": "deepseek",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": true,
    "source": "benchlm",
    "as_of": "2026-10-13",
    "note": "Tracker links a Hugging Face weights repo; licence not verified by us."
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-08-13",
    "source": "benchlm",
    "as_of": "2026-10-13",
    "note": "Aggregator date; matches the 0813 snapshot name."
   },
   "context_tokens": {
    "value": 1000000,
    "source": "ds_pricing",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": 384000,
    "source": "ds_pricing",
    "as_of": "2026-10-13"
   },
   "price_in": {
    "value": 1.32,
    "source": "ds_pricing",
    "as_of": "2026-10-13",
    "note": "Peak rate, cache miss. Off-peak is $0.66."
   },
   "price_out": {
    "value": 3.96,
    "source": "ds_pricing",
    "as_of": "2026-10-13",
    "note": "Peak rate. Off-peak is $1.98."
   },
   "aa_index": {
    "value": 36,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.67,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 95,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 1.66,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "kimi-k3",
   "name": "Kimi K3",
   "company": "moonshot",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": true,
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "Kimi K3 License (commercial use with restrictions)",
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-07-16",
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 1048576,
    "source": "kimi_pricing",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 3.0,
    "source": "kimi_pricing",
    "as_of": "2026-10-13",
    "note": "Cache miss. Cache hit is $0.30."
   },
   "price_out": {
    "value": 15.0,
    "source": "kimi_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 44,
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 2.0,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 43.0,
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 3.14,
    "source": "aa_k3",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "kimi-k2-7-code",
   "name": "Kimi K2.7 Code",
   "company": "moonshot",
   "tier": "mid",
   "aa_setting": "default",
   "open_weights": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "context_tokens": {
    "value": 262144,
    "source": "kimi_pricing",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.95,
    "source": "kimi_pricing",
    "as_of": "2026-10-13",
    "note": "Cache miss."
   },
   "price_out": {
    "value": 4.0,
    "source": "kimi_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 26,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.54,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 75,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 2.8,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "qwen3-8-max",
   "name": "Qwen3.8 Max",
   "company": "alibaba",
   "tier": "flagship",
   "aa_setting": "reasoning",
   "open_weights": {
    "value": false,
    "source": "aa_qmax",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-02",
    "source": "aa_qmax",
    "as_of": "2026-10-13",
    "note": "Artificial Analysis dates the 0902 snapshot 2 Sep; the BenchLM tracker dates Qwen3.8 Max 3 Aug — sources disagree."
   },
   "context_tokens": {
    "value": 984000,
    "source": "aa_qmax",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 2.0,
    "source": "ali_pricing",
    "as_of": "2026-10-13",
    "note": "International (Singapore) list price."
   },
   "price_out": {
    "value": 6.0,
    "source": "ali_pricing",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 45,
    "source": "aa_qmax",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 5.41,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 40.6,
    "source": "aa_qmax",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 2.9,
    "source": "aa_qmax",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "qwen3-8-27b",
   "name": "Qwen3.8 27B",
   "company": "alibaba",
   "tier": "open",
   "aa_setting": "xhigh",
   "open_weights": {
    "value": true,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "Apache 2.0",
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-08-14",
    "source": "aa_q27",
    "as_of": "2026-10-13",
    "note": "BenchLM tracker says 5 Aug — sources disagree."
   },
   "context_tokens": {
    "value": 256000,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.5,
    "source": "aa_q27",
    "as_of": "2026-10-13",
    "note": "Alibaba hosted-API price tracked by Artificial Analysis."
   },
   "price_out": {
    "value": 3.0,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 34,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 1.01,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 45.9,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 3.79,
    "source": "aa_q27",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "nova-2-pro-preview",
   "name": "Nova 2.0 Pro (Preview)",
   "company": "amazon",
   "tier": "flagship",
   "aa_setting": "medium",
   "open_weights": {
    "value": false,
    "source": "aa_nova",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2025-11-27",
    "source": "aa_nova",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 256000,
    "source": "aa_nova",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 1.375,
    "source": "aa_nova",
    "as_of": "2026-10-13",
    "note": "AWS's own pricing tables did not render for our fetch; figure is from Artificial Analysis."
   },
   "price_out": {
    "value": 11.0,
    "source": "aa_nova",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 14,
    "source": "aa_lb",
    "as_of": "2026-10-13",
    "note": "ESTIMATE — Artificial Analysis marks this score as an estimate pending independent evaluation."
   },
   "aa_cost_per_task": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "speed_tps": {
    "value": 122,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 13.15,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": [],
   "estimated": true
  },
  {
   "id": "glm-5-3",
   "name": "GLM-5.3",
   "company": "zai",
   "tier": "flagship",
   "aa_setting": "max",
   "open_weights": {
    "value": true,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "GLM-5.3 License (commercial use with restrictions)",
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-08-18",
    "source": "aa_glm",
    "as_of": "2026-10-13",
    "note": "BenchLM tracker says 14 Aug — sources disagree."
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 1.4,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 4.4,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 45,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 2.01,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 66.9,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 3.32,
    "source": "aa_glm",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "mimo-v2-6-pro",
   "name": "MiMo-V2.6-Pro",
   "company": "xiaomi",
   "tier": "flagship",
   "aa_setting": "reasoning",
   "open_weights": {
    "value": true,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": "MIT",
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-21",
    "source": "aa_mimo",
    "as_of": "2026-10-13",
    "note": "BenchLM tracker says 22 Sep — sources disagree by one day."
   },
   "context_tokens": {
    "value": 1000000,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.435,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 0.87,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 46,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.13,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 47.9,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 3.74,
    "source": "aa_mimo",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "mercury-2-5",
   "name": "Mercury 2.5",
   "company": "inception",
   "tier": "small",
   "aa_setting": "reasoning",
   "open_weights": {
    "value": false,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": "2026-09-08",
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "context_tokens": {
    "value": 260000,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": 0.25,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "price_out": {
    "value": 0.75,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "aa_index": {
    "value": 12,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.12,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 705.8,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 3.38,
    "source": "aa_mercury",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  },
  {
   "id": "celeris-1",
   "name": "Celeris-1",
   "company": "celeris",
   "tier": "small",
   "aa_setting": "default",
   "open_weights": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "license": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "modalities_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "released": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "context_tokens": {
    "value": 131000,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "max_output_tokens": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_in": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "price_out": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "aa_index": {
    "value": 6,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "aa_cost_per_task": {
    "value": 0.05,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "speed_tps": {
    "value": 1522,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "ttft_s": {
    "value": 0.59,
    "source": "aa_lb",
    "as_of": "2026-10-13"
   },
   "terminal_bench_4": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "gdpval_aa_elo": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "deepswe_v1_1": {
    "value": null,
    "source": null,
    "as_of": null
   },
   "swe_bench_verified": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "swebench.com returned no table to our fetch on 2026-10-13; aggregator figures not used"
   },
   "arena_elo": {
    "value": null,
    "source": null,
    "as_of": null,
    "note": "arena.ai/leaderboard returned HTTP 403 to our fetch on 2026-10-13"
   },
   "variants": [],
   "notes": []
  }
 ],
 "voice_models": [
  {
   "name": "GPT-Live-1 (Astra, medium)",
   "company": "openai",
   "sts_index": 83.2,
   "agentic_tau_voice": 74.5,
   "big_bench_audio": 90,
   "tfa_s": 1.34,
   "cost_per_hour": 5.83,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  },
  {
   "name": "Gemini 3.8 Live Extended Thinking (High)",
   "company": "google",
   "sts_index": 82.6,
   "agentic_tau_voice": 68.6,
   "big_bench_audio": 98,
   "tfa_s": 1.35,
   "cost_per_hour": 3.5,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  },
  {
   "name": "Grok Voice Think Fast 2.0 High",
   "company": "xai",
   "sts_index": 81.3,
   "agentic_tau_voice": 56.5,
   "big_bench_audio": 97,
   "tfa_s": 0.7,
   "cost_per_hour": 4.8,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  },
  {
   "name": "GPT-Live-1 (Sol, low)",
   "company": "openai",
   "sts_index": 80.1,
   "agentic_tau_voice": 59.3,
   "big_bench_audio": 89,
   "tfa_s": 1.24,
   "cost_per_hour": 4.47,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  },
  {
   "name": "Gemini 3.8 Live",
   "company": "google",
   "sts_index": 76.0,
   "agentic_tau_voice": 30.1,
   "big_bench_audio": 92,
   "tfa_s": 1.18,
   "cost_per_hour": 0.84,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  },
  {
   "name": "GPT-Realtime-2.1 High",
   "company": "openai",
   "sts_index": 75.2,
   "agentic_tau_voice": 45.7,
   "big_bench_audio": 96,
   "tfa_s": 1.21,
   "cost_per_hour": 10.75,
   "source": "aa_sts",
   "as_of": "2026-10-13"
  }
 ]
}