benchmarks.site

Rankings

Which models come out on top?

A quick overview built from the public leaderboards collected on this site and from our own test tables. It counts how often each model finishes first or in the top 3, and ranks models per task by their average position.

How to read this

Position score (0–100): 100 = first place on every leaderboard where the model appears, 50 = middle of the table on average. Only models on at least 2 leaderboards in a task are ranked.

Take it with a grain of salt

Leaderboards test different things, list different models and update at different times; model names are matched approximately (e.g. dated versions merged). Always open the underlying leaderboard before deciding.

Medal table

Across 47 public leaderboards + 20 of our own result tables
1Claude Fable 5.2111619SimpleBench, LiveBench, FrontierMath, SWE Atlas…
2Gemini 4 Carbon61118CritPt, MathArena, Vending-Bench 2, APEX-Agents…
3Claude Opus 5.54921Terminal-Bench, GDPval-AA, Design Arena, Scale Fortress
4GPT-6 Astra31120Humanity's Last Exam (HLE), ARC-AGI (ARC Prize), EuroEval
5Claude Opus 53519EQ-Bench, SWE-Bench Pro, OSWorld 2.0
6DeepSeek V4.1 Flash2313OpenRouter Rankings, Language models (our test)
7Kimi K3239MCPMark, OCR (our test)
8DeepSeek-V4-Flash236Language models (our test), Language models (our test)
9Qwen3.5-397B-A17B235OCR (our test), OCR (our test)
10Qwen3-ASR-1.7B233ASR (our test), ASR (our test)
11gemini-3.1-flash-lite224ASR (our test), OCR (our test)
12GPT-6.1 Sol1519MultiChallenge
13Kimi K2.6136OCR (our test)
14DeepSeek-V4-Pro136Reasoning level (reasoning_effort) (our test)
15Claude Fable 51214SWE-rebench
16GPT-5.6 Sol1214Context Arena (MRCR)
17GLM 5.3 Flash127Toolathlon (Tool Decathlon)
18OpenAI gpt-live-transcribe122ASR (our test)
19Qwen3-VL 235B-A22B115OCR (our test)
20Anthropic Sonnet 5114F5 Labs CASI & ARS Leaderboard
21gpt-transcribe112ASR (our test)
22Muse-Glimmer-30B112CPTU-Bench
23Qwen3-Embedding-0.6B112Embeddings (our test)
24Mistral-Small-3.2-24B-Instruct112OCR (our test)
25Intelligence-Indeed Agent111OSWorld (OSWorld-Verified)

Best per task

Read text from photos & documents

  1. Kimi K2.6922 boards
  2. Qwen3.5-397B-A17B874 boards
  3. Gemini 3.1 Pro752 boards
  4. MiMo V2.6 Pro692 boards
  5. Mistral-Small-3.2-24B-Instruct572 boards
  6. Qwen3-VL 235B-A22B575 boards
  7. DeepSeek V4.1 Flash494 boards
  8. Qwen3.5-35B-A3B463 boards
  9. Ling-3.0-flash-VL414 boards
  10. Qwen3.8-27B374 boards

All read text from photos & documents tests β†’

Turn speech into text (and back)

  1. Qwen3-ASR-1.7B933 boards
  2. OpenAI gpt-live-transcribe752 boards
  3. whisper-large-v3-turbo732 boards
  4. Soniox stt-async-v6732 boards
  5. whisper-large-v3664 boards
  6. gpt-transcribe582 boards
  7. OpenAI gpt-realtime-whisper542 boards
  8. Qwen3-ASR-0.6B392 boards

All turn speech into text (and back) tests β†’

Chat, answer & summarize

  1. Claude Fable 5.2936 boards
  2. Muse Wonder922 boards
  3. GPT-6 Astra926 boards
  4. Gemini 4 Carbon896 boards
  5. Claude Opus 5.5896 boards
  6. Gemini 4 Argon806 boards
  7. GPT-5.5802 boards
  8. Claude Fable 5.1786 boards
  9. Gemini 3.8 Flash733 boards
  10. GPT-6.1 Sol737 boards

All chat, answer & summarize tests β†’

Write and fix code

  1. Claude Fable 5.2982 boards
  2. Claude Opus 5.5982 boards
  3. Gemini 4 Carbon902 boards
  4. Claude Fable 5.1863 boards
  5. Claude Opus 5833 boards
  6. GPT-6 Astra813 boards
  7. Gemini 4 Argon793 boards
  8. GPT-6.1 Sol782 boards
  9. Muse Wonder 1.0732 boards
  10. Claude Fable 5722 boards

All write and fix code tests β†’

Agents that use tools & computers

  1. Claude Opus 5.5982 boards
  2. Kimi K3932 boards
  3. GPT-6 Astra912 boards
  4. Gemini 4 Carbon833 boards
  5. Claude Fable 5.2803 boards
  6. Claude Fable 5793 boards
  7. Claude Fable 5.1772 boards
  8. GPT-5.6 Sol712 boards
  9. Gemini 4 Argon702 boards
  10. Muse Wonder Full662 boards

All agents that use tools & computers tests β†’

Math & science

  1. GPT-6.1 Sol922 boards
  2. Gemini 4 Carbon902 boards
  3. Claude Opus 5.5752 boards
  4. Gemini 4 Argon642 boards
  5. Kimi K3.1 Pro312 boards

All math & science tests β†’

Polish & other languages

  1. Qwen3.6-27B932 boards
  2. Qwen3-ASR-1.7B933 boards
  3. Kimi K2.6922 boards
  4. GPT-6.1 Sol882 boards
  5. Qwen3.5-397B-A17B874 boards
  6. Qwen3-Embedding-4B862 boards
  7. gemini-3.1-flash-lite853 boards
  8. GPT-6 Astra822 boards
  9. Gemini 3.7 Flash822 boards
  10. Qwen3-Embedding-8B792 boards

All polish & other languages tests β†’

Safety & security

  1. Claude Haiku 5.5892 boards
  2. Claude Opus 5.5652 boards
  3. GPT-5.4612 boards
  4. Claude Opus 4.5502 boards
  5. GPT-6 Luna452 boards
  6. Claude Opus 4.6442 boards
  7. Muse Spark 1.1442 boards
  8. GPT-5.5362 boards

All safety & security tests β†’

Overall rankings

  1. Claude Fable 5.21005 boards
  2. Gemini 4 Carbon874 boards
  3. Claude Opus 5.5835 boards
  4. GPT-6 Astra815 boards
  5. Claude Sonnet 5.5764 boards
  6. GPT-6.1 Terra713 boards
  7. GPT-6.1 Sol684 boards
  8. GPT-6 Sol642 boards
  9. Gemini 4 Argon605 boards
  10. GPT-5.6 Sol602 boards

All overall rankings tests β†’