The Model Index
Every number, sourced

Data & method

The index fits 265 models' results on 9 third-party tests onto one scale, with GPT-4 at 100. Where each number comes from, what counts towards the index and why, and how accurate it is.

252major releases tracked
215linked to an announcement
142dates checked on the lab's own page
69/69lab pages read, 18 Nov 2026

Use the datafree, with credit

Free to use, also from code. Every file below can be downloaded or fetched from any site. The index scores, the release list and the country figures worked out here are ours, under CC BY 4.0: use them for anything, crediting The Model Index with a link. Prices, context windows and each test's results come from the sources listed below and keep their terms.

Cite this. Name the date of the data, so a figure can be found again.

Tladi, T. (2026). The Model Index: AI model capability, prices and releases (Version 2.0) [Data set]. Data of 18 November 2026. https://themodelindex.org/data/
@misc{tladi2026modelindex,
  author       = {Tladi, Thato},
  title        = {The Model Index: AI model capability, prices and releases},
  year         = {2026},
  version      = {2.0},
  howpublished = {\url{https://themodelindex.org/data/}},
  note         = {Data of 2026-11-18. Licensed CC BY 4.0}
}

How the index worksv2.0

The rule. A result counts only if it is all of these:

  • Independent. Someone other than the model's lab ran or verified it.
  • Same conditions for every model. One published setup; self-submitted leaderboards don't count.
  • Checkable. Per-model results are published.
  • Hard to game. Private or refreshed questions where possible.
  • Measures correctness. Graded against answers, tests or task success, not preference.

Results a lab reported about its own model are removed wherever they appear (337 of them). Every test, the tester behind it and the reason it is in or out are listed below.

Testers. Epoch AI's own runs and the third-party results it collects, plus Vals AI, LiveBench, ARC Prize and Scale AI read directly. When two places publish the same tester's results, they count once.

Core tests. Shown on every profile: APEX-Agents, ARC-AGI-2, FrontierMath (tiers 1-3), GPQA Diamond (Epoch AI), Humanity's Last Exam, METR time horizon, SWE-bench Verified, Finance Agent (v2), Terminal-Bench 4.0. Every other test that passes the rule also informs the score, so less-tested models can still be placed.

Model. An item response model estimates one general ability per model from every test it has taken, allowing for how hard and how informative each test is: a test that disagrees with the rest counts for less. It is a single score, not an average of areas. GPT-4 (March 2023) = 100; 50 points is one unit of ability. Pro and Deep Think modes are scored as their own entry, not as the base model.

No tester outweighs the rest. On any one model, a tester's results count as at most 2 tests, however many it runs. The cap is the strictest one that costs no accuracy in the daily check below.

Rated and provisional. A model is rated, and ranked, once it has results in three areas from at least two third-party testers. With less, its score is shown as provisional and it is not ranked. A model with no third-party results has no score.

Uncertainty. Each score has a 90% band from the fit, measured against GPT-4. It assumes the test parameters are right, so treat it as a lower bound on the real uncertainty.

World affordability. On the World page, a $20-a-month plan is set against the median household living standard per person from the latest World Bank household survey for each country, revalued from 2021 purchasing-power dollars to recent US dollars (PPP × local inflation since 2021 ÷ official exchange rate), assuming real living standards unchanged since the survey. "Over a tenth" is the share of people below $2,400 a year on the same basis. Surveys older than 2015, and countries whose official exchange rate is not what people can get, are left out. The 10% line is our threshold, not a test of who can pay. The method was reviewed by a second model (GPT-6 Astra) and its corrections applied.

Sources

Test resultsEpoch AI benchmarking hub ↗
Test resultsVals AI ↗
Test resultsLiveBench ↗
Test resultsARC Prize ↗
Test resultsScale AI ↗
Task lengthMETR time horizons ↗
PricesLabs' own pricing pages
Prices, contextOpenRouter model list ↗
Speed, cost per taskArtificial Analysis ↗
ReleasesLabs' own announcements

A monitor reads 69 official lab pages every 30 minutes and checks the public leaderboards for anything missed. Clear launches on a lab's own site are added automatically; anything uncertain waits for review.

How accurate it ischecked every day

Results it was not given4.3 pointstypical error (median); a fifth of all results hidden at a time
Testers it was not given5.4 pointstypical error predicting a tester from the others
Tester cap2 teststhe strictest cap that costs no accuracy
TesterTests countedTypical missModels compared
METR11.9 pts off36
Tencent Hunyuan / Fudan12.8 pts off19
LiveBench73.6 pts off45
Redwood Research23.8 pts off176
FutureSearch15.5 pts off23
BALROG team15.7 pts off33
Proximal15.8 pts off25
GSO team16.2 pts off30
Vals AI336.2 pts off133
Havard T. Ihle16.3 pts off129
Scale AI26.5 pts off51
Chase Brower16.6 pts off26
SimpleBench team16.9 pts off106
Will Patrick17.1 pts off14
Mercor17.2 pts off62
Stanford CRFM27.2 pts off49
Terminal-Bench team17.4 pts off33
ARC Prize38.1 pts off76
Datacurve18.8 pts off26
Epoch AI149.3 pts off208
Aider110.0 pts off48
Fiction.live110.4 pts off52
GeoBench team110.8 pts off44
Yann Henon111.7 pts off26

Error is in percentage points on each test's own 0–100 scale. "Hidden testers": for every model two or more testers have results for, one tester's results are hidden, the index is refitted from the others, and its prediction is compared with what that tester actually found. Each model and tester pair counts once, however many tests the tester ran.

What counts80 tests · 24 testers

Vals AIBioMysteryBench, Code Migration, CorpFin v2, CyberBench v1.1, EMB, Finance Agent (v2), GPQA Diamond (Vals AI), HLAB-Vals-1.1, IOI, Legal Research Bench, LegalBench, LiveCodeBench, MATH 500, MedCode, MedScribe, MGSM, MMLU Pro, MMMU Pro, MortgageTax, MysteryMechanism, ProgramBench, ProofBench v1.1, Public Benefits Bench v1.1, SAGE, SkillsBench, SRE Bench, SWE-bench, Tax Agent Bench, TaxEval v2, Terminal-Bench 4.0, Terminal-Bench-Science, Vibe Code Bench 1-100, Vibe Code Bench v1.1
Epoch AIAIME-style maths (OTIS mock), Chess puzzles, EBR-bench, FrontierMath (2025 set), FrontierMath (tier 4), FrontierMath (tiers 1-3), FrontierMath tier 4 (2025 set), Furniture assembly, GPQA Diamond (Epoch AI), MATH (level 5), MirrorCode, Mystery game puzzles, SimpleQA Verified, SWE-bench Verified
LiveBenchLiveBench Agentic Coding, LiveBench Coding, LiveBench Data Analysis, LiveBench IF, LiveBench Language, LiveBench Mathematics, LiveBench Reasoning
ARC PrizeARC-AGI-1, ARC-AGI-2, ARC-AGI-3
Redwood ResearchDTBench, LMCA
Scale AIHumanity's Last Exam, SWE-bench Pro (public set)
Stanford CRFMGSM8K (HELM runs), MMLU (HELM runs)
AiderAider Polyglot
BALROG teamBALROG
Chase BrowerVPCT
DatacurveDeepSWE
Fiction.liveFiction.LiveBench
FutureSearchDeepResearch Bench
GeoBench teamGeoBench
GSO teamGSO-Bench
Havard T. IhleWeirdML
MercorAPEX-Agents
METRMETR time horizon
ProximalFrontierSWE
SimpleBench teamSimpleBench
Tencent Hunyuan / FudanCL-bench
Terminal-Bench teamTerminal-Bench 2.0 (Terminus 2 runs)
Will PatrickCadEval
Yann HenonSurface Evolver Bench

Every test was checked against the rule; list last reviewed 2026-10-05.

Left out, and why22 tests

ANLIAll 9 rows come from the Phi-3 report; lab-reported.
ARC AI2All rows are numbers from developers' papers (Qwen, Falcon, Llama 2, Phi, etc.); no independent runs. Old and saturated.
ARC-AGI-1 (Epoch's copy)ARC Prize's own results are read directly from its leaderboard; Epoch's copy would count them twice.
ARC-AGI-2 (Epoch's copy)ARC Prize's own results are read directly from its leaderboard; Epoch's copy would count them twice.
BBHAll rows are developer papers (Qwen, Yi, Baichuan, Llama 2, DeepSeek, Nemotron, Gemma, Phi). Public and saturated.
FrontierCodeEach model is scored in its own vendor harness (claude-code, codex, grok-build, devin, chisel, cursor-cli), so harness and model are confounded. Cognition also makes models on the board (SWE-1.6, 1.7, 2) and the Devin harness. Tasks private; grading is a mix of tests and maintainer rubrics. Only a mini-swe-agent subset (open-weight models only) would be same-conditions, and it has no frontier anchor.
GDPvalCreated and graded by OpenAI (blinded expert pairwise vs a human deliverable), so the tester is a lab with models on the board; only 11 rows, last mid-model-era (GPT-5.2). Pairwise vs human, not an absolute answer key. Judgment call: independence is the main failure; grading by experts is otherwise sound.
HellaSwagIn-index rows are all developer-paper numbers (HELM rows exist in the file but none are in the index). Public, saturated.
LAMBADARows from Qwen and Falcon reports and similar papers; lab-reported, saturated.
Lech Mazur WritingScores are LLM-panel judgments of story quality on a 10-element creative brief; there is no right answer, so it is closer to preference than correctness. Frozen V1 leaderboard (Aug 2025). Judgment call; the rubric exists, so it could be argued either way.
OSWorldEach leaderboard row is a team's own agent plus model, with step budgets varying (15/50/100); several rows are open agents with no named base model and one is from an Anthropic announcement. Public static 369-task set. Only 8 rows are in the index.
OpenBookQARows from developers' papers; lab-reported, saturated.
PIQARows from developers' papers; lab-reported, saturated.
Poker Agent (Vals AI)Scored against the other models at the table, so the opposition changes with who is entered: not the same conditions for every model.
ProofBench (Epoch's copy)Vals AI's own ProofBench results are read directly from Vals; Epoch's copy would count them twice.
Public Benefits Bench v1 (Vals AI)Superseded by its newer version.
Remote Labor IndexScale/CAIS run it with human expert graders on a private 230-project set (good), but agents differ per row (OpenHands CLI, computer-use, vendor agents like Codex or CoWork) and the board reports each agent's best environment. Row names confirm a mix. Could be reconsidered if Scale states a single harness.
ScienceQARows are submissions by model builders to the ScienceQA board or numbers from papers (Phi-3, Falcon2, VisionLLM v2); no common harness or independent tester. Only 6 rows in the index.
Terminal-Bench 2.0 (Vals AI run)Superseded by Terminal-Bench 2.1 and 4.0.
TriviaQARows from Llama 2, Gemma, Phi-3 and similar papers plus one PapersWithCode row; lab-reported.
Voice Code Bench (Vals AI)Mixes speech recognition into the result; out of scope for a text capability index.
WinograndeRows from developers' papers; lab-reported, saturated.

337 results a lab reported about its own model were also removed. Tests with fewer than 15 models are set aside until more are tested, because their difficulty can't be placed reliably: CL-bench Life (14), CUA-bench (3), Cybench (12), ExploitBench (9), OSWorld 2.0 (9), PostTrainBench (11), The Agent Company (11), Time Horizon Index: KSP (13).

Every release252 rows

DateLabReleaseTypeSource
17 Nov 2026OpenAIGPT-6.2auto17 Nov 2026 · OpenAI · flagship · autoflagshipopenai.com ↗
17 Nov 2026OpenAIGPT-6.2 Astraauto17 Nov 2026 · OpenAI · flagship · autoflagshipopenai.com ↗
12 Nov 2026GoogleGemini 4 Hydrogenauto12 Nov 2026 · Google · flagship · autoflagshipblog.google ↗
9 Nov 2026AnthropicClaude Mythos 5.2auto9 Nov 2026 · Anthropic · flagship · autoflagshipanthropic.com ↗
9 Nov 2026AnthropicClaude Fable 5.2auto9 Nov 2026 · Anthropic · flagship · autoflagshipanthropic.com ↗
5 Nov 2026Thinking MachinesInkling 2auto5 Nov 2026 · Thinking Machines · open-weights · autoopen-weightsthinkingmachines.ai ↗
28 Oct 2026GoogleGemini 4 Carbonauto28 Oct 2026 · Google · flagship · autoflagshipblog.google ↗
26 Oct 2026OpenAIGPT-6auto26 Oct 2026 · OpenAI · flagship · autoflagshipopenai.com ↗
26 Oct 2026OpenAIGPT-6.1 Terraauto26 Oct 2026 · OpenAI · flagship · autoflagshipopenai.com ↗
12 Oct 2026MetaMuse Wonder + Muse Wonder Full12 Oct 2026 · Meta · flagshipflagshipabout.fb.com ✓ ↗
7 Oct 2026AnthropicClaude Haiku 5.5auto7 Oct 2026 · Anthropic · small · autosmallanthropic.com ↗
6 Oct 2026MistralMistral Large 4auto6 Oct 2026 · Mistral · open-weights · autoopen-weightsmistral.ai ↗
30 Sep 2026GoogleGemini 4 Argonauto30 Sep 2026 · Google · flagship · autoflagshipblog.google ↗
29 Sep 2026OpenAIGPT-6.1 Sol29 Sep 2026 · OpenAI · flagshipflagshipopenai.com ↗
28 Sep 2026AnthropicClaude Sonnet 5.528 Sep 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
22 Sep 2026OpenAIGPT-6 Sol + GPT-6 Luna22 Sep 2026 · OpenAI · flagshipflagshipopenai.com ↗
22 Sep 2026AnthropicClaude Opus 5.522 Sep 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
21 Sep 2026xAIGrok 4.721 Sep 2026 · xAI · flagshipflagshipx.ai ✓ ↗
21 Sep 2026XiaomiMiMo-V2.6-Pro21 Sep 2026 · Xiaomi · open-weightsopen-weightshuggingface.co ✓ ↗
21 Sep 2026XiaomiMiMo-V2.6-Flash21 Sep 2026 · Xiaomi · open-weightsopen-weightshuggingface.co ✓ ↗
18 Sep 2026StepFunStep 5 Preview18 Sep 2026 · StepFun · flagshipflagshipplatform.stepfun.ai ↗
11 Sep 2026MoonshotKimi K2.8 Preview11 Sep 2026 · Moonshot · flagshipflagship–
10 Sep 2026DeepSeekDeepSeek-V4.1-Flash10 Sep 2026 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
4 Sep 2026OpenAIGPT-6 Astra (preview 09-03)4 Sep 2026 · OpenAI · flagshipflagshipopenai.com ↗
2 Sep 2026MetaMuse Spark 1.32 Sep 2026 · Meta · flagshipflagship–
2 Sep 2026GoogleGemini 3.8 Flash2 Sep 2026 · Google · flagshipflagshipblog.google ✓ ↗
1 Sep 2026AnthropicClaude Fable 5.1 + Mythos 5.11 Sep 2026 · Anthropic · flagshipflagshipanthropic.com ↗
28 Aug 2026TencentHy4 preview28 Aug 2026 · Tencent · open-weightsopen-weightssimonwillison.net ↗
26 Aug 2026AlibabaQwen3.8-Flash26 Aug 2026 · Alibaba · smallsmallqwenlm.github.io ↗
24 Aug 2026AlibabaQwen3.8-Flash-Next24 Aug 2026 · Alibaba · open-weightsopen-weightshuggingface.co ✓ ↗
21 Aug 2026DeepSeekDeepSeek-V4-Flash-Vision-Exp (weights 08-31)21 Aug 2026 · DeepSeek · open-weightsopen-weightshuggingface.co ↗
20 Aug 2026ZhipuGLM-5.3-Flash20 Aug 2026 · Zhipu · open-weightsopen-weightshuggingface.co ↗
14 Aug 2026ZhipuGLM-5.314 Aug 2026 · Zhipu · open-weightsopen-weightsz.ai ✓ ↗
14 Aug 2026AlibabaQwen3.8-27B14 Aug 2026 · Alibaba · open-weightsopen-weightsqwen.ai ↗
13 Aug 2026GoogleGemini 3.7 Flash13 Aug 2026 · Google · flagshipflagshipblog.google ✓ ↗
13 Aug 2026DeepSeekDeepSeek-V4-Pro (GA)13 Aug 2026 · DeepSeek · flagshipflagshipapi-docs.deepseek.com ↗
12 Aug 2026xAIGrok 4.612 Aug 2026 · xAI · flagshipflagshipx.ai ✓ ↗
12 Aug 2026ByteDanceSeed 2.112 Aug 2026 · ByteDance · flagshipflagship–
12 Aug 2026AlibabaQwen3.8-2.4T-A95B12 Aug 2026 · Alibaba · open-weightsopen-weightshuggingface.co ↗
10 Aug 2026MetaMuse Glimmer 30B10 Aug 2026 · Meta · open-weightsopen-weights–
5 Aug 2026MetaMuse Spark 1.2 (+ Muse Code)5 Aug 2026 · Meta · flagshipflagship–
3 Aug 2026AlibabaQwen3.8-Max (preview 07-19; weights 08-12)3 Aug 2026 · Alibaba · flagshipflagshipqwenlm.github.io ↗
31 Jul 2026DeepSeekDeepSeek-V4-Flash (official release; 0731 weights)31 Jul 2026 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ✓ ↗
30 Jul 2026Thinking MachinesInkling-Small30 Jul 2026 · Thinking Machines · open-weightsopen-weightsthinkingmachines.ai ✓ ↗
24 Jul 2026AnthropicClaude Opus 524 Jul 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
23 Jul 2026Ant GroupLing 3.023 Jul 2026 · Ant Group · open-weightsopen-weights–
21 Jul 2026GoogleGemini 3.6 Flash + 3.5 Flash-Lite21 Jul 2026 · Google · flagshipflagshipblog.google ✓ ↗
20 Jul 2026MeituanLongCat 2.020 Jul 2026 · Meituan · open-weightsopen-weights–
16 Jul 2026MoonshotKimi K3 (weights 07-27)16 Jul 2026 · Moonshot · open-weightsopen-weightskimi.com ✓ ↗
15 Jul 2026Thinking MachinesInkling15 Jul 2026 · Thinking Machines · open-weightsopen-weightsthinkingmachines.ai ✓ ↗
9 Jul 2026OpenAIGPT-5.6 (Sol/Terra/Luna; preview 06-26)9 Jul 2026 · OpenAI · flagshipflagshipopenai.com ↗
9 Jul 2026MetaMuse Spark 1.1 (Meta Model API preview)9 Jul 2026 · Meta · flagshipflagshipabout.fb.com ↗
8 Jul 2026xAIGrok 4.58 Jul 2026 · xAI · flagshipflagshipx.ai ↗
30 Jun 2026AnthropicClaude Sonnet 530 Jun 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
16 Jun 2026ZhipuGLM-5.216 Jun 2026 · Zhipu · open-weightsopen-weightsz.ai ✓ ↗
12 Jun 2026MoonshotKimi K2.7 Code12 Jun 2026 · Moonshot · open-weightsopen-weightskimi.com ✓ ↗
9 Jun 2026AnthropicClaude Fable 5 + Mythos 5 (restricted)9 Jun 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
4 Jun 2026NVIDIANemotron 3 Ultra4 Jun 2026 · NVIDIA · open-weightsopen-weightsdeveloper.nvidia.com ✓ ↗
1 Jun 2026MiniMaxMiniMax-M31 Jun 2026 · MiniMax · open-weightsopen-weightshuggingface.co ✓ ↗
1 Jun 2026AlibabaQwen3.7-Plus1 Jun 2026 · Alibaba · flagshipflagship–
29 May 2026StepFunStep 3.7 Flash29 May 2026 · StepFun · open-weightsopen-weightshuggingface.co ↗
28 May 2026AnthropicClaude Opus 4.828 May 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
20 May 2026xAIGrok Build 0.1 (coding)20 May 2026 · xAI · reasoningreasoningx.ai ↗
19 May 2026GoogleGemini 3.5 Flash (3.5 Pro announced; unreleased)19 May 2026 · Google · flagshipflagshipblog.google ✓ ↗
19 May 2026AlibabaQwen3.7-Max19 May 2026 · Alibaba · flagshipflagship–
28 Apr 2026MistralMistral Medium 3.528 Apr 2026 · Mistral · open-weightsopen-weightsmistral.ai ↗
24 Apr 2026DeepSeekDeepSeek-V4 preview (V4-Pro/V4-Flash)24 Apr 2026 · DeepSeek · open-weightsopen-weights–
23 Apr 2026TencentHy3 preview23 Apr 2026 · Tencent · open-weightsopen-weightshuggingface.co ✓ ↗
23 Apr 2026OpenAIGPT-5.5 (Thinking/Pro)23 Apr 2026 · OpenAI · flagshipflagshipopenai.com ✓ ↗
22 Apr 2026XiaomiMiMo-V2.5-Pro22 Apr 2026 · Xiaomi · open-weightsopen-weightshuggingface.co ↗
20 Apr 2026MoonshotKimi K2.620 Apr 2026 · Moonshot · open-weightsopen-weightskimi.com ✓ ↗
20 Apr 2026AlibabaQwen3.6-Max-Preview20 Apr 2026 · Alibaba · flagshipflagship–
17 Apr 2026xAIGrok 4.3 (beta; API/GA 04-30)17 Apr 2026 · xAI · flagshipflagshipdocs.x.ai ↗
16 Apr 2026AnthropicClaude Opus 4.716 Apr 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
16 Apr 2026AlibabaQwen3.6-35B-A3B (open)16 Apr 2026 · Alibaba · open-weightsopen-weights–
8 Apr 2026MetaMuse Spark (proprietary; MSL)8 Apr 2026 · Meta · flagshipflagshipabout.fb.com ✓ ↗
7 Apr 2026ZhipuGLM-5.17 Apr 2026 · Zhipu · open-weightsopen-weightshuggingface.co ↗
7 Apr 2026AnthropicClaude Mythos Preview (restricted; Project Glasswing)7 Apr 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
2 Apr 2026GoogleGemma 42 Apr 2026 · Google · open-weightsopen-weightsblog.google ✓ ↗
2 Apr 2026AlibabaQwen3.6-Plus2 Apr 2026 · Alibaba · flagshipflagship–
18 Mar 2026MiniMaxMiniMax-M2.718 Mar 2026 · MiniMax · open-weightsopen-weightsminimax.io ✓ ↗
16 Mar 2026MistralMistral Small 416 Mar 2026 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
11 Mar 2026NVIDIANemotron 3 Super11 Mar 2026 · NVIDIA · open-weightsopen-weightsblogs.nvidia.com ✓ ↗
5 Mar 2026OpenAIGPT-5.4 (Thinking/Pro; mini/nano 03-17)5 Mar 2026 · OpenAI · flagshipflagshipopenai.com ✓ ↗
3 Mar 2026GoogleGemini 3.1 Flash-Lite3 Mar 2026 · Google · smallsmallblog.google ✓ ↗
19 Feb 2026GoogleGemini 3.1 Pro19 Feb 2026 · Google · flagshipflagshipblog.google ✓ ↗
17 Feb 2026xAIGrok 4.20 (beta; GA/API 03-10)17 Feb 2026 · xAI · flagshipflagshipdocs.x.ai ↗
17 Feb 2026AnthropicClaude Sonnet 4.617 Feb 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
16 Feb 2026AlibabaQwen3.5 (397B open + 3.5-Plus)16 Feb 2026 · Alibaba · open-weightsopen-weightsqwenlm.github.io ↗
14 Feb 2026ByteDanceSeed 2.014 Feb 2026 · ByteDance · flagshipflagshipseed.bytedance.com ✓ ↗
12 Feb 2026OpenAIGPT-5.3-Codex-Spark12 Feb 2026 · OpenAI · smallsmallopenai.com ✓ ↗
12 Feb 2026MiniMaxMiniMax-M2.512 Feb 2026 · MiniMax · open-weightsopen-weightshuggingface.co ✓ ↗
11 Feb 2026ZhipuGLM-511 Feb 2026 · Zhipu · open-weightsopen-weightshuggingface.co ↗
5 Feb 2026OpenAIGPT-5.3-Codex5 Feb 2026 · OpenAI · reasoningreasoningopenai.com ↗
5 Feb 2026AnthropicClaude Opus 4.65 Feb 2026 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
2 Feb 2026StepFunStep 3.5 Flash2 Feb 2026 · StepFun · open-weightsopen-weightshuggingface.co ✓ ↗
27 Jan 2026MoonshotKimi K2.527 Jan 2026 · Moonshot · open-weightsopen-weightskimi.com ↗
26 Jan 2026AlibabaQwen3-Max-Thinking26 Jan 2026 · Alibaba · reasoningreasoningqwen.ai ↗
23 Dec 2025MiniMaxMiniMax-M2.123 Dec 2025 · MiniMax · open-weightsopen-weightshuggingface.co ↗
22 Dec 2025ZhipuGLM-4.722 Dec 2025 · Zhipu · open-weightsopen-weightshuggingface.co ✓ ↗
18 Dec 2025OpenAIGPT-5.2-Codex18 Dec 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
17 Dec 2025GoogleGemini 3 Flash17 Dec 2025 · Google · smallsmallblog.google ✓ ↗
16 Dec 2025XiaomiMiMo-V2-Flash16 Dec 2025 · Xiaomi · open-weightsopen-weightsmimo.xiaomi.com ✓ ↗
15 Dec 2025NVIDIANemotron 3 Nano15 Dec 2025 · NVIDIA · open-weightsopen-weightsnvidianews.nvidia.com ✓ ↗
11 Dec 2025OpenAIGPT-5.211 Dec 2025 · OpenAI · flagshipflagshipopenai.com ↗
9 Dec 2025MistralDevstral 2 (+ Devstral Small 2)9 Dec 2025 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
4 Dec 2025GoogleGemini 3 Deep Think4 Dec 2025 · Google · reasoningreasoningblog.google ✓ ↗
2 Dec 2025MistralMistral Large 3 (+ Ministral 3)2 Dec 2025 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
2 Dec 2025AmazonNova 22 Dec 2025 · Amazon · flagshipflagshipaboutamazon.com ✓ ↗
1 Dec 2025DeepSeekDeepSeek-V3.2 (+ Speciale)1 Dec 2025 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
27 Nov 2025DeepSeekDeepSeekMath-V227 Nov 2025 · DeepSeek · open-weightsopen-weights–
24 Nov 2025AnthropicClaude Opus 4.524 Nov 2025 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
19 Nov 2025xAIGrok 4.1 Fast (+ Agent Tools API)19 Nov 2025 · xAI · smallsmallx.ai ✓ ↗
19 Nov 2025OpenAIGPT-5.1-Codex-Max19 Nov 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
18 Nov 2025GoogleGemini 3 Pro18 Nov 2025 · Google · flagshipflagshipblog.google ✓ ↗
17 Nov 2025xAIGrok 4.117 Nov 2025 · xAI · flagshipflagshipx.ai ✓ ↗
13 Nov 2025BaiduERNIE 5.013 Nov 2025 · Baidu · flagshipflagship–
12 Nov 2025OpenAIGPT-5.112 Nov 2025 · OpenAI · flagshipflagshipopenai.com ✓ ↗
6 Nov 2025MoonshotKimi K2 Thinking6 Nov 2025 · Moonshot · reasoningreasoningkimi.com ↗
30 Oct 2025MoonshotKimi Linear 48B-A3B30 Oct 2025 · Moonshot · open-weightsopen-weightshuggingface.co ↗
27 Oct 2025MiniMaxMiniMax-M227 Oct 2025 · MiniMax · open-weightsopen-weights–
15 Oct 2025AnthropicClaude Haiku 4.515 Oct 2025 · Anthropic · smallsmallanthropic.com ✓ ↗
14 Oct 2025Ant GroupRing-1T14 Oct 2025 · Ant Group · open-weightsopen-weightshuggingface.co ↗
9 Oct 2025Ant GroupLing-1T9 Oct 2025 · Ant Group · open-weightsopen-weightshuggingface.co ↗
30 Sep 2025ZhipuGLM-4.630 Sep 2025 · Zhipu · open-weightsopen-weightshuggingface.co ✓ ↗
29 Sep 2025DeepSeekDeepSeek-V3.2-Exp (sparse attention)29 Sep 2025 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
29 Sep 2025AnthropicClaude Sonnet 4.529 Sep 2025 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
19 Sep 2025xAIGrok 4 Fast19 Sep 2025 · xAI · smallsmallx.ai ✓ ↗
15 Sep 2025OpenAIGPT-5-Codex15 Sep 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
11 Sep 2025AlibabaQwen3-Next-80B-A3B11 Sep 2025 · Alibaba · open-weightsopen-weights–
5 Sep 2025AlibabaQwen3-Max (preview; official 09-24)5 Sep 2025 · Alibaba · flagshipflagship–
1 Sep 2025MeituanLongCat-Flash1 Sep 2025 · Meituan · open-weightsopen-weightshuggingface.co ↗
28 Aug 2025xAIGrok Code Fast 128 Aug 2025 · xAI · smallsmallx.ai ✓ ↗
28 Aug 2025MicrosoftMAI-1-preview28 Aug 2025 · Microsoft · flagshipflagship–
23 Aug 2025xAIGrok 2.5 open weights23 Aug 2025 · xAI · open-weightsopen-weightstechcrunch.com ↗
21 Aug 2025DeepSeekDeepSeek-V3.1 (hybrid reasoning)21 Aug 2025 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
7 Aug 2025OpenAIGPT-57 Aug 2025 · OpenAI · flagshipflagshipopenai.com ✓ ↗
5 Aug 2025OpenAIgpt-oss-120b/20b5 Aug 2025 · OpenAI · open-weightsopen-weightsopenai.com ✓ ↗
5 Aug 2025AnthropicClaude Opus 4.15 Aug 2025 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
1 Aug 2025GoogleGemini 2.5 Deep Think1 Aug 2025 · Google · reasoningreasoningblog.google ✓ ↗
28 Jul 2025ZhipuGLM-4.528 Jul 2025 · Zhipu · open-weightsopen-weightsz.ai ✓ ↗
25 Jul 2025StepFunStep 325 Jul 2025 · StepFun · open-weightsopen-weights–
22 Jul 2025AlibabaQwen3-Coder22 Jul 2025 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
11 Jul 2025MoonshotKimi K211 Jul 2025 · Moonshot · open-weightsopen-weightskimi.com ↗
9 Jul 2025xAIGrok 4 (+ Heavy)9 Jul 2025 · xAI · flagshipflagshipx.ai ✓ ↗
27 Jun 2025TencentHunyuan-A13B27 Jun 2025 · Tencent · open-weightsopen-weightshuggingface.co ✓ ↗
17 Jun 2025GoogleGemini 2.5 Flash-Lite (preview; 2.5 Pro/Flash GA)17 Jun 2025 · Google · smallsmalldevelopers.googleblog.com ✓ ↗
16 Jun 2025MiniMaxMiniMax-M116 Jun 2025 · MiniMax · open-weightsopen-weightshuggingface.co ✓ ↗
11 Jun 2025ByteDanceSeed 1.611 Jun 2025 · ByteDance · reasoningreasoning–
10 Jun 2025OpenAIo3-pro10 Jun 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
10 Jun 2025MistralMagistral (Small/Medium)10 Jun 2025 · Mistral · reasoningreasoningmistral.ai ✓ ↗
28 May 2025DeepSeekDeepSeek-R1-052828 May 2025 · DeepSeek · reasoningreasoningapi-docs.deepseek.com ↗
22 May 2025AnthropicClaude Opus 4 + Sonnet 422 May 2025 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
21 May 2025MistralDevstral21 May 2025 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
20 May 2025GoogleGemma 3n (preview; full 06-26)20 May 2025 · Google · smallsmall–
16 May 2025OpenAIcodex-1 (Codex research preview)16 May 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
7 May 2025MistralMistral Medium 37 May 2025 · Mistral · flagshipflagshipmistral.ai ✓ ↗
30 Apr 2025XiaomiMiMo-7B30 Apr 2025 · Xiaomi · open-weightsopen-weightshuggingface.co ✓ ↗
30 Apr 2025MicrosoftPhi-4-reasoning30 Apr 2025 · Microsoft · open-weightsopen-weightshuggingface.co ✓ ↗
30 Apr 2025AmazonNova Premier30 Apr 2025 · Amazon · flagshipflagshipaws.amazon.com ✓ ↗
29 Apr 2025AlibabaQwen329 Apr 2025 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
17 Apr 2025GoogleGemini 2.5 Flash17 Apr 2025 · Google · smallsmallblog.google ✓ ↗
16 Apr 2025OpenAIo3 + o4-mini16 Apr 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
14 Apr 2025OpenAIGPT-4.114 Apr 2025 · OpenAI · flagshipflagshipen.wikipedia.org ↗
5 Apr 2025MetaLlama 4 (Scout/Maverick)5 Apr 2025 · Meta · open-weightsopen-weightsabout.fb.com ✓ ↗
25 Mar 2025GoogleGemini 2.5 Pro25 Mar 2025 · Google · reasoningreasoningblog.google ✓ ↗
21 Mar 2025TencentHunyuan-T121 Mar 2025 · Tencent · reasoningreasoning–
16 Mar 2025BaiduERNIE X116 Mar 2025 · Baidu · reasoningreasoning–
16 Mar 2025BaiduERNIE 4.516 Mar 2025 · Baidu · flagshipflagship–
12 Mar 2025GoogleGemma 312 Mar 2025 · Google · open-weightsopen-weightsblog.google ✓ ↗
6 Mar 2025AlibabaQwQ-32B6 Mar 2025 · Alibaba · reasoningreasoningqwenlm.github.io ✓ ↗
27 Feb 2025OpenAIGPT-4.527 Feb 2025 · OpenAI · flagshipflagshipen.wikipedia.org ↗
24 Feb 2025AnthropicClaude 3.7 Sonnet24 Feb 2025 · Anthropic · reasoningreasoninganthropic.com ✓ ↗
17 Feb 2025xAIGrok 3 (+ Grok 3 mini)17 Feb 2025 · xAI · flagshipflagshipx.ai ✓ ↗
5 Feb 2025GoogleGemini 2.0 Pro (exp) + 2.0 Flash GA5 Feb 2025 · Google · flagshipflagshipblog.google ✓ ↗
31 Jan 2025OpenAIo3-mini31 Jan 2025 · OpenAI · reasoningreasoningopenai.com ✓ ↗
30 Jan 2025MistralMistral Small 330 Jan 2025 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
28 Jan 2025AlibabaQwen2.5-Max28 Jan 2025 · Alibaba · flagshipflagshipqwenlm.github.io ✓ ↗
26 Jan 2025AlibabaQwen2.5-VL26 Jan 2025 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
22 Jan 2025ByteDanceDoubao 1.5 Pro22 Jan 2025 · ByteDance · flagshipflagship–
20 Jan 2025MoonshotKimi k1.520 Jan 2025 · Moonshot · reasoningreasoningkimi.com ✓ ↗
20 Jan 2025DeepSeekDeepSeek-R120 Jan 2025 · DeepSeek · reasoningreasoningapi-docs.deepseek.com ↗
15 Jan 2025MiniMaxMiniMax-0115 Jan 2025 · MiniMax · open-weightsopen-weightshuggingface.co ↗
26 Dec 2024DeepSeekDeepSeek-V326 Dec 2024 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
20 Dec 2024OpenAIo3 / o3-mini (announced; preview only)20 Dec 2024 · OpenAI · reasoningreasoningen.wikipedia.org ↗
19 Dec 2024GoogleGemini 2.0 Flash Thinking (exp)19 Dec 2024 · Google · reasoningreasoningai.google.dev ✓ ↗
12 Dec 2024MicrosoftPhi-412 Dec 2024 · Microsoft · open-weightsopen-weights–
11 Dec 2024GoogleGemini 2.0 Flash (experimental)11 Dec 2024 · Google · flagshipflagshipblog.google ✓ ↗
6 Dec 2024MetaLlama 3.3 70B6 Dec 2024 · Meta · open-weightsopen-weightsabout.fb.com ↗
5 Dec 2024OpenAIo1 (full) + o1 pro5 Dec 2024 · OpenAI · reasoningreasoning–
3 Dec 2024AmazonAmazon Nova3 Dec 2024 · Amazon · flagshipflagship–
28 Nov 2024AlibabaQwQ-32B-Preview28 Nov 2024 · Alibaba · reasoningreasoningqwen.ai ↗
20 Nov 2024DeepSeekDeepSeek-R1-Lite-Preview20 Nov 2024 · DeepSeek · reasoningreasoningdeepseek.com ✓ ↗
12 Nov 2024AlibabaQwen2.5-Coder-32B12 Nov 2024 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
5 Nov 2024TencentHunyuan-Large5 Nov 2024 · Tencent · open-weightsopen-weights–
22 Oct 2024AnthropicClaude 3.5 Sonnet (new) + 3.5 Haiku (avail. 11-04)22 Oct 2024 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
16 Oct 2024MistralMinistral 3B/8B16 Oct 2024 · Mistral · smallsmallmistral.ai ✓ ↗
15 Oct 2024NVIDIALlama-3.1-Nemotron-70B15 Oct 2024 · NVIDIA · open-weightsopen-weights–
25 Sep 2024MetaLlama 3.2 (1B/3B text + 11B/90B vision)25 Sep 2024 · Meta · open-weightsopen-weightsai.meta.com ✓ ↗
19 Sep 2024AlibabaQwen2.519 Sep 2024 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
12 Sep 2024OpenAIo1-preview / o1-mini12 Sep 2024 · OpenAI · reasoningreasoningopenai.com ✓ ↗
11 Sep 2024MistralPixtral 12B11 Sep 2024 · Mistral · open-weightsopen-weightsmistral.ai ↗
13 Aug 2024xAIGrok-213 Aug 2024 · xAI · flagshipflagshipx.ai ✓ ↗
24 Jul 2024MistralMistral Large 224 Jul 2024 · Mistral · flagshipflagshipmistral.ai ✓ ↗
23 Jul 2024MetaLlama 3.1 (405B)23 Jul 2024 · Meta · open-weightsopen-weightsai.meta.com ✓ ↗
18 Jul 2024OpenAIGPT-4o mini18 Jul 2024 · OpenAI · smallsmalldevelopers.openai.com ↗
27 Jun 2024GoogleGemma 227 Jun 2024 · Google · open-weightsopen-weightsblog.google ✓ ↗
20 Jun 2024AnthropicClaude 3.5 Sonnet20 Jun 2024 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
17 Jun 2024DeepSeekDeepSeek-Coder-V217 Jun 2024 · DeepSeek · open-weightsopen-weightsapi-docs.deepseek.com ↗
14 Jun 2024NVIDIANemotron-4 340B14 Jun 2024 · NVIDIA · open-weightsopen-weightsblogs.nvidia.com ✓ ↗
7 Jun 2024AlibabaQwen27 Jun 2024 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
29 May 2024MistralCodestral29 May 2024 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
15 May 2024ByteDanceDoubao Pro15 May 2024 · ByteDance · flagshipflagship–
14 May 2024GoogleGemini 1.5 Flash14 May 2024 · Google · smallsmallblog.google ✓ ↗
13 May 2024OpenAIGPT-4o13 May 2024 · OpenAI · flagshipflagshipopenai.com ✓ ↗
6 May 2024DeepSeekDeepSeek-V26 May 2024 · DeepSeek · open-weightsopen-weightsgithub.com ✓ ↗
23 Apr 2024MicrosoftPhi-323 Apr 2024 · Microsoft · open-weightsopen-weightshuggingface.co ✓ ↗
18 Apr 2024MetaLlama 318 Apr 2024 · Meta · open-weightsopen-weightsai.meta.com ✓ ↗
10 Apr 2024MistralMixtral 8x22B10 Apr 2024 · Mistral · open-weightsopen-weightsmistral.ai ↗
28 Mar 2024xAIGrok-1.528 Mar 2024 · xAI · flagshipflagshipx.ai ✓ ↗
17 Mar 2024xAIGrok-1 open weights17 Mar 2024 · xAI · open-weightsopen-weightsx.ai ✓ ↗
4 Mar 2024AnthropicClaude 3 (Opus/Sonnet; Haiku 03-13)4 Mar 2024 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
26 Feb 2024MistralMistral Large26 Feb 2024 · Mistral · flagshipflagshipmistral.ai ✓ ↗
21 Feb 2024GoogleGemma (2B/7B)21 Feb 2024 · Google · open-weightsopen-weightsblog.google ✓ ↗
15 Feb 2024GoogleGemini 1.5 Pro15 Feb 2024 · Google · flagshipflagshipblog.google ✓ ↗
8 Feb 2024GoogleGemini 1.0 Ultra (Gemini Advanced)8 Feb 2024 · Google · flagshipflagshipblog.google ↗
4 Feb 2024AlibabaQwen1.54 Feb 2024 · Alibaba · open-weightsopen-weightsqwenlm.github.io ✓ ↗
16 Jan 2024ZhipuGLM-416 Jan 2024 · Zhipu · flagshipflagship–
12 Dec 2023MicrosoftPhi-212 Dec 2023 · Microsoft · open-weightsopen-weightshuggingface.co ✓ ↗
8 Dec 2023MistralMixtral 8x7B8 Dec 2023 · Mistral · open-weightsopen-weightsmistral.ai ↗
6 Dec 2023GoogleGemini 1.0 (Pro/Nano; Ultra announced)6 Dec 2023 · Google · flagshipflagshipblog.google ✓ ↗
30 Nov 2023AlibabaQwen-72B30 Nov 2023 · Alibaba · open-weightsopen-weightsgithub.com ✓ ↗
29 Nov 2023DeepSeekDeepSeek LLM 67B29 Nov 2023 · DeepSeek · open-weightsopen-weights–
21 Nov 2023AnthropicClaude 2.121 Nov 2023 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
6 Nov 2023OpenAIGPT-4 Turbo6 Nov 2023 · OpenAI · flagshipflagshipdevelopers.openai.com ↗
4 Nov 2023xAIGrok-14 Nov 2023 · xAI · flagshipflagshipen.wikipedia.org ↗
2 Nov 2023DeepSeekDeepSeek Coder2 Nov 2023 · DeepSeek · open-weightsopen-weights–
17 Oct 2023BaiduERNIE 4.017 Oct 2023 · Baidu · flagshipflagship–
9 Oct 2023MoonshotKimi Chat (debut; 200K-char context)9 Oct 2023 · Moonshot · flagshipflagshipkimi.com ↗
27 Sep 2023MistralMistral 7B27 Sep 2023 · Mistral · open-weightsopen-weightsmistral.ai ✓ ↗
24 Aug 2023MetaCode Llama24 Aug 2023 · Meta · open-weightsopen-weightsai.meta.com ✓ ↗
9 Aug 2023AnthropicClaude Instant 1.29 Aug 2023 · Anthropic · smallsmallanthropic.com ✓ ↗
3 Aug 2023AlibabaQwen-7B (first open weights)3 Aug 2023 · Alibaba · open-weightsopen-weights–
18 Jul 2023MetaLlama 218 Jul 2023 · Meta · open-weightsopen-weightsai.meta.com ✓ ↗
11 Jul 2023AnthropicClaude 211 Jul 2023 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
10 May 2023GooglePaLM 210 May 2023 · Google · flagshipflagshipblog.google ✓ ↗
11 Apr 2023AlibabaQwen / Tongyi Qianwen (beta)11 Apr 2023 · Alibaba · flagshipflagshipen.wikipedia.org ↗
14 Mar 2023ZhipuChatGLM-6B14 Mar 2023 · Zhipu · open-weightsopen-weightshuggingface.co ✓ ↗
14 Mar 2023OpenAIGPT-414 Mar 2023 · OpenAI · flagshipflagshipopenai.com ✓ ↗
14 Mar 2023AnthropicClaude (1)14 Mar 2023 · Anthropic · flagshipflagshipanthropic.com ✓ ↗
24 Feb 2023MetaLLaMA (1)24 Feb 2023 · Meta · open-weightsopen-weightsen.wikipedia.org ↗
30 Nov 2022OpenAIChatGPT (GPT-3.5)30 Nov 2022 · OpenAI · flagshipflagshipopenai.com ✓ ↗