Model Markets
Get Free API KeyAPI

Rankings

Intelligence, cost, and their combined efficiency for current available models.

Intelligence

46 eligible · v2.5 · Sep 14, 2026 · Higher is better

  1. 99.0Claude Fable 5.1
  2. 95.6Claude Fable 5
  3. 94.1Claude Opus 5
  4. 87.7Gemini 3.7 Flash
  5. 85.3GPT-5.6 Sol
  6. 84.4GPT-6 Astra
  7. 83.9GPT-5.5
  8. 82.0*Gemini 3.8 Flash
  9. 79.7Gemini 3.1 Pro
  10. 79.2*Qwen3.8 Max
  11. 79.0Claude Opus 4.7
  12. 76.2Claude Opus 4.6
  13. 74.9GPT-5.6 Terra
  14. 72.8GPT-5.4
  15. 72.1Gemini 3.5 Flash
  16. 70.4*GLM 5.3
  17. 70.3Kimi K3
  18. 70.1Claude Opus 4.8
  19. 66.4Gemini 3.6 Flash
  20. 65.0DeepSeek V4 Pro 0813
  21. 64.1Grok 4.6
  22. 63.4*Grok 4.20 Multi Agent
  23. 61.9*GLM 5.2
  24. 60.3Claude Sonnet 4.6
  25. 60.1Grok 4.5
  26. 60.1*Claude Sonnet 5
  27. 55.4*Claude Opus 4.5
  28. 55.3*DeepSeek V4 Pro
  29. 54.6*GLM 5.3 Flash
  30. 54.0*Qwen3.8 27B
  31. 52.7GPT-5.6 Luna
  32. 50.5*Kimi K2.6
  33. 49.3GPT-5.2
  34. 45.4Inkling
  35. 32.6*MiniMax M3
  36. 27.7*Claude Sonnet 4.5
  37. 25.7*GPT-5.1
  38. 20.2GPT-5.4 Mini
  39. 17.6Gemini 3.5 Flash Lite
  40. 16.6*Grok 4.3
  41. 15.0*GPT-5
  42. 14.7GPT-5.4 Nano
  43. 9.0*Claude Haiku 4.5
  44. 3.7*GPT-4.1
  45. 0.0*GPT-5 Mini
  46. 0.0*GPT-5 Nano
Available core inputs from LiveBench overall, ARC-AGI-2 verified, and LMArena Text overall are equally weighted. * marks a provisional two-source score. Bar hover text and accessible labels include the missing source and possible full-core range. Methodology.

Cost

44 eligible · output cost per LiveBench case · Lower is better

  1. $0.0020Grok Build 0.1
  2. $0.0087GLM 5.3 Flash
  3. $0.017MiniMax M3
  4. $0.021Qwen3.8 Flash
  5. $0.022DeepSeek V4.1 Flash
  6. $0.026GPT-5.6 Luna
  7. $0.028Grok 4.3
  8. $0.029Gemini 3.5 Flash Lite
  9. $0.042Kimi K2.7 Code
  10. $0.054DeepSeek V4 Pro
  11. $0.055Gemini 3.6 Flash
  12. $0.056GLM 5.2
  13. $0.057Qwen3.7 Max
  14. $0.058GPT-5.4 Nano
  15. $0.063Grok 4.5
  16. $0.072Qwen3.8 27B
  17. $0.085Gemini 3.7 Flash
  18. $0.090Inkling
  19. $0.092NVIDIA Nemotron 3 Ultra 550B A55B BF16
  20. $0.094DeepSeek V4 Pro 0813
  21. $0.095Kimi K2.6
  22. $0.097Grok 4.6
  23. $0.107Qwen3.8 Max
  24. $0.117GPT-5.6 Sol
  25. $0.140Gemini 3.5 Flash
  26. $0.160Gemini 3.8 Flash
  27. $0.161Gemini 3.1 Pro
  28. $0.161GPT-5.2
  29. $0.194Claude Sonnet 5
  30. $0.194Kimi K3
  31. $0.198Claude Sonnet 4.6
  32. $0.214GPT-5.4 Mini
  33. $0.241Claude Opus 4.6
  34. $0.248GLM 5.3
  35. $0.266GPT-5.6 Terra
  36. $0.274GPT-5.4
  37. $0.286Claude Opus 4.7
  38. $0.341GPT-5.5
  39. $0.371Claude Opus 5
  40. $0.380Claude Opus 4.5
  41. $0.520GPT-6 Astra
  42. $0.604Claude Opus 4.8
  43. $0.824Claude Fable 5.1
  44. $1.01Claude Fable 5
Published output tokens per LiveBench case multiplied by the lowest current sourced USD output-token rate.

Efficiency

38 eligible · v1.5 · Higher is better

  1. 77.3*GLM 5.3 Flash
  2. 69.9Gemini 3.7 Flash
  3. 65.3GPT-5.6 Sol
  4. 64.8GPT-5.6 Luna
  5. 63.7Gemini 3.6 Flash
  6. 63.2*Qwen3.8 Max
  7. 61.3*GLM 5.2
  8. 60.4*Gemini 3.8 Flash
  9. 59.2Gemini 3.1 Pro
  10. 59.2*MiniMax M3
  11. 59.2Grok 4.5
  12. 58.4*DeepSeek V4 Pro
  13. 57.6Claude Opus 5
  14. 57.4DeepSeek V4 Pro 0813
  15. 56.8Gemini 3.5 Flash
  16. 56.7Grok 4.6
  17. 54.8*Qwen3.8 27B
  18. 53.4GPT-5.5
  19. 53.2Claude Opus 4.6
  20. 52.7Claude Opus 4.7
  21. 52.5Kimi K3
  22. 51.7Claude Fable 5.1
  23. 51.5GPT-5.6 Terra
  24. 50.1*Kimi K2.6
  25. 50.1GPT-5.4
  26. 50.0*GLM 5.3
  27. 49.2GPT-6 Astra
  28. 48.1Inkling
  29. 47.8Claude Fable 5
  30. 47.4*Claude Sonnet 5
  31. 47.3Claude Sonnet 4.6
  32. 46.2Gemini 3.5 Flash Lite
  33. 46.0*Grok 4.3
  34. 44.0GPT-5.2
  35. 40.5Claude Opus 4.8
  36. 38.0*Claude Opus 4.5
  37. 37.5GPT-5.4 Nano
  38. 26.5GPT-5.4 Mini
Equal halves: MM Relative Intelligence and logarithmically normalized affordability across the current available common cohort. * marks provisional Intelligence coverage; the possible full-core range carries through to Efficiency.

Every panel includes the current available eligible cohort in exact rank order. Intelligence admits labeled two-of-three partial core scores; unavailable benchmark, token, or pricing inputs are never estimated.

Send Feedback