AnotherWrapper
    Build with the APIs
    LLMImageVideoSpeechTranscription

    Reasoning

    🏆 Leaderboard

    As of August 21, 2026, GPT-5.6 Sol is #1 for reasoning at 94.6%. Ranked by GPQA: graduate-level science questions that resist simple search. Best AI for reasoning, ranked by Humanity's Last Exam, GPQA Diamond, and ARC-AGI. Compare HLE vs GPQA next to API price.

    Updated August 21, 2026282 models33 providers
    GPT-5.6 Sol
    OpenAI · Proprietary
    		94.6%82.8%—71.6%74.5%92.5%64.8%7.8%$35.00
    Claude Mythos Preview
    Anthropic · Proprietary
    		94.6%—64.7%—————$60.00
    Claude Mythos 5
    Anthropic · Proprietary
    		94.6%94.6%——————$60.00
    Gemini 3.1 Pro
    Google · Proprietary
    		94.3%94.4%51.4%77.3%98%77.1%79.6%0.4%$17.50
    Claude Opus 4.7
    Anthropic · Proprietary
    		94.2%86.4%54.7%50.6%93.5%75.8%62.9%—$30.00
    GPT-5.5
    OpenAI · Proprietary
    		93.6%77.3%52.2%—76.2%85%69%0.4%$35.00
    Claude Opus 4.8
    Anthropic · Proprietary
    		93.6%85.3%57.9%39.5%88%72.1%64.8%1.5%$30.00
    Kimi K3OSS
    Moonshot AI · Open Source
    		93.5%93.5%56%42.7%94.5%60.4%——$18.00
    GPT-5.2 Pro
    OpenAI · Proprietary
    		93.2%—36.6%—81.2%54.2%57.4%—$189.00
    Grok 4.5
    xAI · Proprietary
    		93%93.4%—53.5%85.7%52.6%70%0.3%$8.00
    GPT-5.6 Terra
    OpenAI · Proprietary
    		92.9%77.3%—43.1%60.2%83.9%48.9%0.8%$14.00
    GPT-5.4
    OpenAI · Proprietary
    		92.8%89.4%39.8%44.8%76.7%73.3%—0.2%$17.50
    Qwen3.8 MaxOSS
    Qwen · Open Source
    		92.6%92.7%43.6%46.3%————$8.00
    Qwen3.7 Max
    Qwen · Proprietary
    		92.4%90.9%41.4%58.5%——70.4%—$5.00
    GPT-5.2
    OpenAI · Proprietary
    		92.4%73.2%34.5%35.4%12.3%52.9%45.8%—$15.75
    GPT-5.6 Luna
    OpenAI · Proprietary
    		92.3%63.6%—41.7%34.2%59.5%46.8%0.2%$1.40
    Gemini 3 Pro
    Google · Proprietary
    		91.9%91.9%45.8%72.1%75%31.1%76.4%—$14.00
    Claude Opus 4.6
    Anthropic · Proprietary
    		91.3%88.4%53.1%41.0%86%68.8%67.6%—$30.00
    GLM-5.2OSS
    Z AI · Open Source
    		91.2%71.2%54.7%38.1%77%22.8%——$5.80
    Kimi K2.6OSS
    Moonshot AI · Open Source
    		90.5%90.8%36.4%38.7%————$4.93
    Qwen3.6 Plus
    Qwen · Proprietary
    		90.4%88.4%28.8%49.1%————$3.50
    Hy3OSS
    Tencent · Open Source
    		90.4%———————$0.66
    Gemini 3 Flash
    Google · Proprietary
    		90.4%89.4%43.5%68.7%21.5%33.6%61.1%—$3.50
    Qwen3.7-Plus
    Qwen · Proprietary
    		90.3%87.9%34.7%—————$1.60
    DeepSeek-V4-Pro-MaxOSS
    DeepSeek · Open Source
    		90.1%89.7%48.2%57.9%————$5.22
    Claude Sonnet 4.6
    Anthropic · Proprietary
    		89.9%78.8%46.8%29%86%58.3%——$18.00
    Inkling-SmallOSS
    Thinking Machines · Open Source · via Thinking Machines Lab
    		89.5%88.5%31.6%19.5%84%40.1%——$1.50
    Qwen3.8-27BOSS
    Qwen · Open Source
    		89.2%—30.8%—————$3.65
    Solar Pro 4
    Upstage · Proprietary
    		89%———————$1.50
    Seed 2.0 Pro
    ByteDance · Proprietary
    		88.9%———————$3.50
    Qwen3.5-397B-A17BOSS
    Qwen · Open Source
    		88.4%85.9%28.7%—————$4.20
    GPT-5.1 Thinking
    OpenAI · Proprietary
    		88.1%—23.7%—————$11.25
    GPT-5.1 Instant
    OpenAI · Proprietary
    		88.1%—6.8%—————$11.25
    GPT-5.1 High
    OpenAI · Proprietary
    		88.1%———72.8%———$11.25
    GPT-5.1
    OpenAI · Proprietary
    		88.1%66.7%23.7%48.9%5.8%17.6%53.2%—$11.25
    GPT-5 Medium
    OpenAI · Proprietary
    		88.1%———56.2%7.5%——$11.25
    DeepSeek-V4-Flash-MaxOSS
    DeepSeek · Open Source
    		88.1%—45.1%34.1%————$0.42
    GPT-5.4 mini
    OpenAI · Proprietary
    		88%88%28.2%28.6%63.7%18.9%——$5.25
    Qwen3.6-27BOSS
    Qwen · Open Source
    		87.8%85.9%24%—————$4.20
    GPT-5.3 Codex
    OpenAI · Proprietary
    		87.7%87.7%——66.7%———$15.75
    Kimi K2.5OSS
    Moonshot AI · Open Source
    		87.6%—50.2%—65.3%11.8%46.8%—$3.68
    Grok-4
    xAI · Proprietary
    		87.5%87.5%40%—66.7%15.9%——$18.00
    DeepSeek-V4-Flash-0423OSS
    DeepSeek · Open Source
    		87.4%—40.3%28.9%————$0.30
    GPT-5 High
    OpenAI · Proprietary
    		87.3%———65.7%9.9%——$11.25
    Nemotron 3 Ultra (550B A55B)OSS
    NVIDIA · Open Source · via OpenRouter
    		87%86.1%37.4%—————$2.70
    Claude Opus 4.5
    Anthropic · Proprietary
    		87%85.5%25.2%41.8%80%37.6%62%—$30.00
    Gemini 3.1 Flash-Lite
    Google · Proprietary
    		86.9%81.8%16%43.3%————$1.75
    Qwen3.5-122B-A10BOSS
    Qwen · Open Source
    		86.6%—47.5%—————$3.60
    Gemini 2.5 Pro Preview 06-05
    Google · Proprietary
    		86.4%84.8%21.6%54%31.3%0%62.4%—$11.25
    GLM-5.1OSS
    Z AI · Open Source
    		86.2%89.9%52.3%37.3%——55.1%—$5.80
    Showing 1–50 of 282 models

    Next step

    You found the model. Now ship the product.

    Auth, billing, and the API layer are already decided. Fork 8 finished AI apps, or have us build the first version with you.

    Build with the APIsHave us build it with you

    The index

    All Large Language Models

    282 models across 33 providers. Search or jump to a lab — every model page stays linked here.

    OpenAI51Anthropic22Google25Cursor2xAI15DeepSeek16MiniMax6Qwen38AI21 Labs2Amazon5Baidu1ByteDance2Cohere3IBM1Inception1inclusionAI1LG AI Research1Meituan4Meta12Microsoft5Mistral23Moonshot AI9Nous Research1NVIDIA5Poolside2Sakana AI1StepFun1Tencent1Thinking Machines1Unisound1Upstage1Xiaomi3Z AI20

    Baidu

    1 models
    ERNIE 4.5

    Inception

    1 models
    Mercury 2

    inclusionAI

    1 models
    Ling-3.0-flash

    LG AI Research

    1 models
    K-EXAONE-236B-A23B

    Nous Research

    1 models
    Hermes 3 70B

    Sakana AI

    1 models
    Sakana Namazu

    StepFun

    1 models
    Step-3.5-Flash

    Tencent

    1 models
    Hy3

    Thinking Machines

    1 models
    Inkling-Small

    Unisound

    1 models
    U2

    Upstage

    1 models
    Solar Pro 4

    What is the best AI for reasoning?

    As of August 21, 2026, GPT-5.6 Sol by OpenAI is #1 for reasoning at 94.6%. Ranked by GPQA: graduate-level science questions that resist simple search. This board also tracks GPQA, GPQA Diamond, HLE, SimpleQA. Next on the same board: Claude Mythos 5 and Claude Mythos Preview. Related leaders: Gemini 3.7 Flash on GPQA Diamond at 94.8%; Claude Opus 5 on HLE at 64.7%. This reasoning leaderboard ranks models by GPQA. Scores come from public evals. Prices are the live API rates in the table above.

    Sources: GPQA (GitHub) (https://github.com/idavidrein/gpqa); OpenAI API pricing (https://developers.openai.com/api/docs/pricing); Anthropic Claude API pricing (https://platform.claude.com/docs/en/about-claude/pricing); Gemini API pricing (https://ai.google.dev/gemini-api/docs/pricing)

    Top 8 for reasoning. Ranked by GPQA: graduate-level science questions that resist simple search. Input and output are dollars per million tokens.
    RankModelGPQAInput /MOutput /M
    1GPT-5.6 Sol94.6%$5.00$30.00
    2Claude Mythos 594.6%$10.00$50.00
    3Claude Mythos Preview94.6%$10.00$50.00
    4Gemini 3.1 Pro94.3%$2.50$15.00
    5Claude Opus 4.794.2%$5.00$25.00
    6Claude Opus 4.893.6%$5.00$25.00
    7GPT-5.593.6%$5.00$30.00
    8Kimi K393.5%$3.00$15.00

    What is Humanity's Last Exam?

    Humanity's Last Exam is the expert-written set people search when they want a reasoning leaderboard. About 2,500 questions. Search engines should not solve them. Nature published the paper in 2026. This page ranks HLE next to GPQA Diamond and ARC-AGI, plus API price.

    HLE vs GPQA Diamond vs ARC-AGI

    HLE is the wide expert exam. GPQA Diamond is graduate science. ARC-AGI is novel puzzles. A model can ace GPQA and stall on ARC-AGI. That is the point of keeping three columns.

    Related LLM leaderboards

    LLM LeaderboardCodingMathAgenticWritingScienceFinanceLong ContextVisionOpen Source

    Individual benchmark pages

    Rank one eval at a time. All LLM benchmarks.

    HLESWE BenchHumanity's Last Exam (No Tools)Humanity's Last Exam (With Tools)Terminal-Bench 2.0Terminal-BenchTerminal-Bench 2.1Terminal-Bench 2Terminal-Bench 3.0ARC-AGI-1 Verified

    Reasoning FAQ

    Who ranks #1 on the Reasoning leaderboard?

    As of August 21, 2026, GPT-5.6 Sol by OpenAI ranks #1 on GPQA at 94.6%. API pricing is $5.00/M input and $30.00/M output.

    What is the best AI for reasoning?

    As of August 21, 2026, GPT-5.6 Sol by OpenAI is #1 for reasoning at 94.6%. Ranked by GPQA: graduate-level science questions that resist simple search. This board also tracks GPQA, GPQA Diamond, HLE, SimpleQA. Next on the same board: Claude Mythos 5 and Claude Mythos Preview. Related leaders: Gemini 3.7 Flash on GPQA Diamond at 94.8%; Claude Opus 5 on HLE at 64.7%.

    What are the top models on GPQA?

    The current GPQA ranking as of August 21, 2026 is 1. GPT-5.6 Sol at 94.6%; 2. Claude Mythos 5 at 94.6%; 3. Claude Mythos Preview at 94.6%.

    Which reasoning model is the cheapest?

    Llama 3.2 3B Instruct is the cheapest scored model on this reasoning leaderboard at $0.01/M input and $0.02/M output ($0.03 blended). GPT-5.6 Sol still leads GPQA at 94.6%.

    Should I always pick the #1 GPQA model?

    Not automatically. GPT-5.6 Sol leads GPQA, but a cheaper scored model can be the better production choice if the quality gap is small. Use the table to weigh GPQA against input/output price, context window, and related evals.

    How often is the Reasoning leaderboard updated?

    Scores and API prices on this page are refreshed from published evals and provider rates. The snapshot is labeled August 21, 2026. Treat it as a current index, not a one-off blog post.

    What is GPQA?

    GPQA is a graduate-level science QA set written to resist simple search. GPQA Diamond is the harder expert-validated subset.

    How is reasoning different from coding evals?

    Reasoning benches score knowledge and puzzle-solving. Coding benches score repository edits and terminal work. A high HLE score does not guarantee a high SWE-bench score.

    What is Humanity's Last Exam?

    Humanity's Last Exam (HLE) is a 2,500-question expert exam from the Center for AI Safety and Scale AI, published in Nature. It is built so that search and memorization fail. HLE with tools and HLE without tools are different scores.

    What is GPQA Diamond?

    GPQA is graduate-level science QA written to resist Google. GPQA Diamond is the harder expert-validated subset. It is the usual public “is this model actually smart” number next to HLE.

    What is ARC-AGI?

    ARC-AGI is François Chollet's Abstraction and Reasoning Corpus. It scores fluid intelligence on novel visual puzzles, not textbook recall. ARC-AGI 2 is the current hard public set. High HLE does not imply high ARC-AGI.

    Is Humanity's Last Exam the same as GPQA?

    No. GPQA Diamond is graduate science questions. HLE is a broader expert exam across fields, designed to stay hard after GPQA started to saturate. Rank both.