Skip to content

Charts

Every chart uses the same data as the leaderboard. Colour shows model type: blue proprietary, orange open weights, green other.

Overall score

Top 25 ranked models with 90% intervals (whiskers).

  1. 1Claude Opus 5.5Anthropic86.7
  2. 2GPT-6.1 SolOpenAI84.2
  3. 3GPT-6 AstraOpenAI84.0
  4. 4Claude Fable 5.1Anthropic83.2
  5. 5Gemini 4 ArgonGoogle82.9
  6. 6Claude Sonnet 5.5Anthropic82.2
  7. 7Claude Fable 5Anthropic81.3
  8. 8Claude Opus 5Anthropic80.1
  9. 9GPT-6 SolOpenAI78.6
  10. 10GPT-5.6 SolOpenAI75.3
  11. 11Muse Wonder FullMeta74.8
  12. 12MiMo-V2.6-ProXiaomi73.7
  13. 13GPT-5.5 ProOpenAI73.5
  14. 14Muse WonderMeta71.5
  15. 15GPT-5.6 TerraOpenAI70.2
  16. 16Claude Opus 4.8Anthropic70.0
  17. 17Kimi K3Moonshot AI70.0
  18. 18Qwen3.8 MaxAlibaba69.7
  19. 19Step 5 PreviewStepFun69.6
  20. 20GPT-5.5OpenAI69.2
  21. 21Gemini 3.8 FlashGoogle69.2
  22. 22GPT-5.4 ProOpenAI68.7
  23. 23Claude Opus 4.7 (Adaptive)Anthropic68.7
  24. 24Ling 3.1 FlashInclusionAI68.1
  25. 25GLM-5.3Z.AI68.0
Higher is better. Whiskers that overlap mean the order is not certain.

Category leaders

Top 8 models in each capability area.

Agentic & tool useFull ranking →

  1. 1GPT-6 Astra86.5
  2. 2Claude Opus 585.2
  3. 3GPT-5.6 Sol83.7
  4. 4Claude Fable 582.9
  5. 5Qwen3.8 Max82.9
  6. 6Ling 3.1 Flash82.3
  7. 7Kimi K381.9
  8. 8GPT-5.581.7
Higher is better.

Instruction followingFull ranking →

  1. 1Grok 4.397.1
  2. 2Qwen3.8 Max94.7
  3. 3MAI-Thinking-194.7
  4. 4GLM-5.193.8
  5. 5GPT-5.2-Codex93.8
  6. 6MiniMax M393.8
  7. 7MiMo-V2.5-Pro93.8
  8. 8GPT-5.593.3
Higher is better.

Score vs price

Blended price per 1M tokens (3:1 input to output). The dashed line is the Pareto frontier: no model is both cheaper and better than a model on it.

Pareto frontierMost attractive quadrant162 models
020406080100$0.01$0.1$1$10$100$1,000Blended price, USD per 1M tokens (log scale) · lower is betterScore · higher is betterGPT-6.1 SolGPT-6 AstraGemini 4 ArgonGPT-6 SolMiMo-V2.6-ProLing 3.1 FlashGLM-5.3-FlashLing 3.0 Flash

Score vs context window

Pareto frontierMost attractive quadrant220 models
025507510010K100K1M10MContext window, tokens (log scale) · higher is betterScore · higher is betterClaude Opus 5.5GPT-6.1 SolClaude Fable 5GPT-6 SolGPT-5.6 SolMiMo-V2.6-ProGemini 3 ProLlama 4 Scout

Open vs proprietary over time

Best score released up to each month. The gap between the lines is how far open weights trail the frontier.

Best proprietaryBest open weights
02550751002023-112024-042024-092025-022025-072025-122026-052026-10Best score so far

Cheapest of the top 40

Blended price per 1M tokens, split into input and output share.

Input shareOutput share
  1. Ling 3.1 Flash$0.11
  2. MiMo-V2.6-Flash$0.18
  3. Claude Haiku 5.5$0.20
  4. GPT-6 Luna$0.20
  5. MiniMax M3.1 Flash$0.26
  6. DeepSeek V4.1 Flash 1017$0.26
  7. GLM-5.3$0.41
  8. GPT-5.6 Luna$0.45
  9. MiMo-V2.6-Pro$0.54
  10. Step 5 Preview$1.43
  11. Gemini 3.8 Flash$1.50
  12. Gemini 3.7 Flash$1.50
  13. Muse Spark 1.1$2.00
  14. Muse Spark 1.2$2.00
  15. Grok 4.6$3.00
  16. GPT-6.1 Sol$4.00
  17. Gemini 4 Argon$4.00
  18. Claude Sonnet 5.5$4.00
  19. GPT-6 Sol$4.00
  20. Claude Sonnet 5$4.00
Lower is better.

Largest context windows

Among ranked models.

  1. 1Llama 4 Scout10M
  2. 2Gemini 3 Pro2M
  3. 3Grok 4.202M
  4. 4Grok 4.1 Fast (Reasoning)2M
  5. 5Grok 4 Fast (Reasoning)2M
  6. 6Gemini 1.5 Pro2M
  7. 7GPT-6.1 Sol1.05M
  8. 8GPT-6 Astra1.05M
  9. 9GPT-6 Sol1.05M
  10. 10GPT-5.6 Sol1.05M
  11. 11GPT-5.6 Terra1.05M
  12. 12Kimi K31.05M
  13. 13GPT-5.4 Pro1.05M
  14. 14GPT-5.41.05M
  15. 15GPT-6 Luna1.05M
  16. 16GPT-5.6 Luna1.05M
  17. 17Claude Opus 5.51M
  18. 18Claude Fable 5.11M
  19. 19Gemini 4 Argon1M
  20. 20Claude Sonnet 5.51M
Higher is better.

Score vs output speed

Our own measurements only.

No speed measurements published yet. See the speed page.