Benchmark Almanac

Terminal-Bench 4.1

Claude Opus 5.5 65.8% leads since November 2026
Not in model cards

The Terminal-Bench 4.0 tasks with hardened verifiers: agents complete 66 hard tasks in containerized terminals across software, ML, science, security, and operations. Test scripts verify each result, scored as resolution rate.

It kept the 4.0 tasks and hardened the verifiers of 44 of them against reward hacking, then regraded the saved 4.0 runs. 162 trials changed score, and every model stayed within its earlier 95% interval.

Released
November 2026
Kind
Terminal tasks
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 13th, 2026
43
Models scored on its own leaderboard

2 models share the top score of 65.8%

Best score per model from the benchmark's own leaderboard, read on November 13th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

43 models. Scroll sideways for the rest.

  1. 65.8 Claude Opus 5.5
  2. 65.8 Claude Fable 5.2
  3. 63.9 Claude Sonnet 5.5
  4. 60.9 GPT-6.1 Sol
  5. 60.3 GPT-6 Astra
  6. 59.7 Kimi K3.1 Pro
  7. 57.6 Claude Fable 5.1
  8. 56.4 Gemini 4 Hydrogen
  9. 53.3 Claude Opus 5
  10. 53.3 GPT-6.1 Terra
  11. 50.3 GPT-6 Sol
  12. 44.9 Claude Fable 5
  13. 41.2 GLM-5.3
  14. 40.0 Claude Haiku 5.5
  15. 38.8 Grok 4.7
  16. 38.5 GPT-5.6 Sol
  17. 38.5 DeepSeek V4.1 Flash 1017
  18. 37.9 Tencent Hy4
  19. 37.0 GLM-5.3-Flash
  20. 30.6 Muse Ember
  21. 27.6 Mistral Large 4
  22. 27.0 Qwen3.8 Max (0902)
  23. 27.0 MiMo-V2.6 Pro
  24. 23.6 Claude Opus 4.8
  25. 23.0 Kimi K3
  26. 21.5 GPT-5.6 Terra
  27. 19.7 Grok 4.6
  28. 18.2 Gemini 3.8 Flash
  29. 17.6 GPT-5.6 Luna
  30. 16.1 MiMo-V2.6 Flash
  31. 15.2 GPT-6 Luna
  32. 14.8 Muse Spark 1.3
  33. 13.0 Inkling 2 Small
  34. 12.4 Grok 4.5
  35. 12.4 Gemini 3.7 Flash
  36. 12.1 Claude Sonnet 5
  37. 7.58 MiniMax-M3
  38. 6.67 DeepSeek V4 Pro 0813
  39. 6.36 Beam 1.0
  40. 1.82 GPT-5.4 Mini
  41. 0.91 Inkling-Small
  42. 0.30 Nemotron 3.5 Super VL
  43. 0.0 Inkling
  1. 1 Claude Opus 5.5 max 65.8%
  2. 2 Claude Fable 5.2 max 65.8%
  3. 3 Claude Sonnet 5.5 max 63.9%
  4. 4 GPT-6.1 Sol max 60.9%
  5. 5 GPT-6 Astra max 60.3%
  6. 6 Kimi K3.1 Pro max 59.7%
  7. 7 Claude Fable 5.1 xhigh 57.6%
  8. 8 Gemini 4 Hydrogen high 56.4%
  9. 9 Claude Opus 5 xhigh 53.3%
  10. 10 GPT-6.1 Terra max 53.3%
  11. 11 GPT-6 Sol max 50.3%
  12. 12 Claude Fable 5 max 44.9%
  13. 13 GLM-5.3 max 41.2%
  14. 14 Claude Haiku 5.5 max 40%
  15. 15 Grok 4.7 xhigh 38.8%
  16. 16 GPT-5.6 Sol max 38.5%
  17. 17 DeepSeek V4.1 Flash 1017 max 38.5%
  18. 18 Tencent Hy4 max 37.9%
  19. 19 GLM-5.3-Flash max 37%
  20. 20 Muse Ember xhigh 30.6%
  21. 21 Mistral Large 4 max 27.6%
  22. 22 Qwen3.8 Max (0902) max 27%
  23. 23 MiMo-V2.6 Pro none 27%
  24. 24 Claude Opus 4.8 max 23.6%
  25. 25 Kimi K3 max 23%
  26. 26 GPT-5.6 Terra max 21.5%
  27. 27 Grok 4.6 high 19.7%
  28. 28 Gemini 3.8 Flash high 18.2%
  29. 29 GPT-5.6 Luna max 17.6%
  30. 30 MiMo-V2.6 Flash none 16.1%
  31. 31 GPT-6 Luna max 15.2%
  32. 32 Muse Spark 1.3 xhigh 14.8%
  33. 33 Inkling 2 Small xhigh 13%
  34. 34 Grok 4.5 high 12.4%
  35. 35 Gemini 3.7 Flash high 12.4%
  36. 36 Claude Sonnet 5 max 12.1%
  37. 37 MiniMax-M3 none 7.58%
  38. 38 DeepSeek V4 Pro 0813 max 6.67%
  39. 39 Beam 1.0 none 6.36%
  40. 40 GPT-5.4 Mini xhigh 1.82%
  41. 41 Inkling-Small none 0.91%
  42. 42 Nemotron 3.5 Super VL none 0.30%
  43. 43 Inkling none 0%

Source: Official Terminal-Bench 4.1 leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5.5 65.8% max, Claude Code Sep 2026 Official leaderboard
2 Claude Fable 5.2 65.8% max, Claude Code Nov 2026 Official leaderboard
3 Claude Sonnet 5.5 63.9% max, Claude Code Sep 2026 Official leaderboard
4 GPT-6.1 Sol 60.9% max, Codex Sep 2026 Official leaderboard
5 GPT-6 Astra 60.3% max, Codex Sep 2026 Official leaderboard
6 Kimi K3.1 Pro 59.7% max, Claude Code Oct 2026 Official leaderboard
7 Claude Fable 5.1 57.6% xhigh, Claude Code Sep 2026 Official leaderboard
8 Gemini 4 Hydrogen 56.4% high, Antigravity CLI Nov 2026 Official leaderboard
9 Claude Opus 5 53.3% xhigh, Claude Code Jul 2026 Official leaderboard
10 GPT-6.1 Terra 53.3% max, Codex Oct 2026 Official leaderboard
11 GPT-6 Sol 50.3% max, Codex Sep 2026 Official leaderboard
12 Claude Fable 5 44.9% max, Claude Code Jun 2026 Official leaderboard
13 GLM-5.3 41.2% max, Claude Code Aug 2026 Official leaderboard
14 Claude Haiku 5.5 40% max, Claude Code Oct 2026 Official leaderboard
15 Grok 4.7 38.8% xhigh, Grok Build Sep 2026 Official leaderboard
16 GPT-5.6 Sol 38.5% max, Codex Jul 2026 Official leaderboard
17 DeepSeek V4.1 Flash 1017 38.5% max, Claude Code Oct 2026 Official leaderboard
18 Tencent Hy4 37.9% max, Claude Code Nov 2026 Official leaderboard
19 GLM-5.3-Flash 37% max, Claude Code Aug 2026 Official leaderboard
20 Muse Ember 30.6% xhigh, Muse Code Nov 2026 Official leaderboard
21 Mistral Large 4 27.6% max, Claude Code Oct 2026 Official leaderboard
22 Qwen3.8 Max (0902) 27% max, Claude Code Sep 2026 Official leaderboard
23 MiMo-V2.6 Pro 27% none, mini-swe-agent Sep 2026 Official leaderboard
24 Claude Opus 4.8 23.6% max, Claude Code May 2026 Official leaderboard
25 Kimi K3 23% max, Claude Code Jul 2026 Official leaderboard
26 GPT-5.6 Terra 21.5% max, Codex Jul 2026 Official leaderboard
27 Grok 4.6 19.7% high, Grok Build Aug 2026 Official leaderboard
28 Gemini 3.8 Flash 18.2% high, mini-SWE-agent Sep 2026 Official leaderboard
29 GPT-5.6 Luna 17.6% max, Codex Jul 2026 Official leaderboard
30 MiMo-V2.6 Flash 16.1% none, mini-swe-agent Sep 2026 Official leaderboard
31 GPT-6 Luna 15.2% max, Codex Sep 2026 Official leaderboard
32 Muse Spark 1.3 14.8% xhigh, Muse Code Sep 2026 Official leaderboard
33 Inkling 2 Small 13% xhigh, mini-swe-agent Nov 2026 Official leaderboard
34 Grok 4.5 12.4% high, Grok Build Jul 2026 Official leaderboard
35 Gemini 3.7 Flash 12.4% high, mini-SWE-agent Aug 2026 Official leaderboard
36 Claude Sonnet 5 12.1% max, Claude Code Jun 2026 Official leaderboard
37 MiniMax-M3 7.58% none, mini-swe-agent Jun 2026 Official leaderboard
38 DeepSeek V4 Pro 0813 6.67% max, Claude Code Aug 2026 Official leaderboard
39 Beam 1.0 6.36% none, mini-swe-agent Oct 2026 Official leaderboard
40 GPT-5.4 Mini 1.82% xhigh, Codex Mar 2026 Official leaderboard
41 Inkling-Small 0.91% none, mini-swe-agent Jul 2026 Official leaderboard
42 Nemotron 3.5 Super VL 0.30% none, mini-swe-agent Oct 2026 Official leaderboard
43 Inkling 0% none, mini-swe-agent Jul 2026 Official leaderboard

Pareto: accuracy against time

Each point is one agent, model and effort. The board plots total run time; this plots the mean time per trial, the same frontier at a fixed trial count. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 80% 1001,00010K Mean time per trial (s), log scale, lower is better Accuracy GPT-5.4 Mini (xhigh): Accuracy 1.82%, mean time per trial 142 s Claude Opus 4.8 (max): Accuracy 23.6%, mean time per trial 5,163 s MiniMax-M3 (none): Accuracy 7.58%, mean time per trial 204 s Claude Fable 5 (max): Accuracy 44.9%, mean time per trial 4,203 s Claude Sonnet 5 (max): Accuracy 12.1%, mean time per trial 6,510 s Grok 4.5 (high): Accuracy 12.4%, mean time per trial 3,162 s GPT-5.6 Luna (max): Accuracy 17.6%, mean time per trial 4,088 s GPT-5.6 Sol (max): Accuracy 38.5%, mean time per trial 2,388 s GPT-5.6 Terra (max): Accuracy 21.5%, mean time per trial 2,515 s Inkling (none): Accuracy 0%, mean time per trial 174 s Inkling-Small (none): Accuracy 0.91%, mean time per trial 141 s Kimi K3 (max): Accuracy 23%, mean time per trial 142 s Claude Opus 5 (xhigh): Accuracy 53.3%, mean time per trial 4,515 s Claude Opus 5 (max): Accuracy 52.1%, mean time per trial 4,793 s Claude Opus 5 (high): Accuracy 49.7%, mean time per trial 3,822 s Claude Opus 5 (medium): Accuracy 45.1%, mean time per trial 3,293 s Claude Opus 5 (low): Accuracy 34.9%, mean time per trial 2,947 s Grok 4.6 (high): Accuracy 19.7%, mean time per trial 2,352 s DeepSeek V4 Pro 0813 (max): Accuracy 6.67%, mean time per trial 138 s Gemini 3.7 Flash (high): Accuracy 12.4%, mean time per trial 1,689 s GLM-5.3 (max): Accuracy 41.2%, mean time per trial 5,831 s GLM-5.3-Flash (none): Accuracy 35.5%, mean time per trial 6,210 s Claude Fable 5.1 (xhigh): Accuracy 57.6%, mean time per trial 3,202 s Claude Fable 5.1 (max): Accuracy 57.3%, mean time per trial 3,894 s Claude Fable 5.1 (high): Accuracy 54.9%, mean time per trial 3,018 s Claude Fable 5.1 (medium): Accuracy 53.9%, mean time per trial 2,528 s Claude Fable 5.1 (low): Accuracy 43.3%, mean time per trial 2,291 s Qwen3.8 Max (0902) (max): Accuracy 27%, mean time per trial 9,953 s Gemini 3.8 Flash (high): Accuracy 18.2%, mean time per trial 1,995 s Muse Spark 1.3 (xhigh): Accuracy 14.8%, mean time per trial 4,342 s GPT-6 Astra (xhigh): Accuracy 59.7%, mean time per trial 2,209 s GPT-6 Astra (medium): Accuracy 55.1%, mean time per trial 1,886 s GPT-6 Astra (low): Accuracy 50.9%, mean time per trial 1,684 s Grok 4.7 (xhigh): Accuracy 38.8%, mean time per trial 5,701 s MiMo-V2.6 Flash (none): Accuracy 16.1%, mean time per trial 161 s MiMo-V2.6 Pro (none): Accuracy 27%, mean time per trial 168 s GPT-6 Luna (max): Accuracy 15.2%, mean time per trial 2,284 s GPT-6 Sol (max): Accuracy 50.3%, mean time per trial 2,662 s Claude Sonnet 5.5 (max): Accuracy 63.9%, mean time per trial 7,432 s Beam 1.0 (none): Accuracy 6.36%, mean time per trial 179 s Mistral Large 4 (max): Accuracy 27.6%, mean time per trial 143 s Claude Haiku 5.5 (max): Accuracy 40%, mean time per trial 141 s Nemotron 3.5 Super VL (none): Accuracy 0.3%, mean time per trial 136 s DeepSeek V4.1 Flash 1017 (max): Accuracy 38.5%, mean time per trial 148 s Tencent Hy4 (max): Accuracy 37.9%, mean time per trial 126 s Inkling 2 Small (xhigh): Accuracy 13%, mean time per trial 7,273 s Claude Fable 5.2 (max): Accuracy 65.8%, mean time per trial 7,333 s Muse Ember (xhigh): Accuracy 30.6%, mean time per trial 2,288 s Gemini 4 Hydrogen (high): Accuracy 56.4%, mean time per trial 1,926 s GLM-5.3-Flash (max): Accuracy 37%, mean time per trial 120 s GPT-6 Astra (max): Accuracy 60.3%, mean time per trial 2,796 s GPT-6 Astra (high): Accuracy 60%, mean time per trial 2,114 s Claude Opus 5.5 (max): Accuracy 65.8%, mean time per trial 4,790 s GPT-6.1 Sol (max): Accuracy 60.9%, mean time per trial 3,177 s Kimi K3.1 Pro (max): Accuracy 59.7%, mean time per trial 123 s GPT-6.1 Terra (max): Accuracy 53.3%, mean time per trial 123 s Claude Opus 5.5 (max)GPT-6.1 Sol (max)GPT-6 Astra (max)GPT-6 Astra (high)Kimi K3.1 Pro (max)GPT-6.1 Terra (max)GLM-5.3-Flash (max) 0% 20% 40% 60% 80% 1001,00010K Mean time per trial (s), log scale, lower is better Accuracy GPT-5.4 Mini (xhigh): Accuracy 1.82%, mean time per trial 142 s Claude Opus 4.8 (max): Accuracy 23.6%, mean time per trial 5,163 s MiniMax-M3 (none): Accuracy 7.58%, mean time per trial 204 s Claude Fable 5 (max): Accuracy 44.9%, mean time per trial 4,203 s Claude Sonnet 5 (max): Accuracy 12.1%, mean time per trial 6,510 s Grok 4.5 (high): Accuracy 12.4%, mean time per trial 3,162 s GPT-5.6 Luna (max): Accuracy 17.6%, mean time per trial 4,088 s GPT-5.6 Sol (max): Accuracy 38.5%, mean time per trial 2,388 s GPT-5.6 Terra (max): Accuracy 21.5%, mean time per trial 2,515 s Inkling (none): Accuracy 0%, mean time per trial 174 s Inkling-Small (none): Accuracy 0.91%, mean time per trial 141 s Kimi K3 (max): Accuracy 23%, mean time per trial 142 s Claude Opus 5 (xhigh): Accuracy 53.3%, mean time per trial 4,515 s Claude Opus 5 (max): Accuracy 52.1%, mean time per trial 4,793 s Claude Opus 5 (high): Accuracy 49.7%, mean time per trial 3,822 s Claude Opus 5 (medium): Accuracy 45.1%, mean time per trial 3,293 s Claude Opus 5 (low): Accuracy 34.9%, mean time per trial 2,947 s Grok 4.6 (high): Accuracy 19.7%, mean time per trial 2,352 s DeepSeek V4 Pro 0813 (max): Accuracy 6.67%, mean time per trial 138 s Gemini 3.7 Flash (high): Accuracy 12.4%, mean time per trial 1,689 s GLM-5.3 (max): Accuracy 41.2%, mean time per trial 5,831 s GLM-5.3-Flash (none): Accuracy 35.5%, mean time per trial 6,210 s Claude Fable 5.1 (xhigh): Accuracy 57.6%, mean time per trial 3,202 s Claude Fable 5.1 (max): Accuracy 57.3%, mean time per trial 3,894 s Claude Fable 5.1 (high): Accuracy 54.9%, mean time per trial 3,018 s Claude Fable 5.1 (medium): Accuracy 53.9%, mean time per trial 2,528 s Claude Fable 5.1 (low): Accuracy 43.3%, mean time per trial 2,291 s Qwen3.8 Max (0902) (max): Accuracy 27%, mean time per trial 9,953 s Gemini 3.8 Flash (high): Accuracy 18.2%, mean time per trial 1,995 s Muse Spark 1.3 (xhigh): Accuracy 14.8%, mean time per trial 4,342 s GPT-6 Astra (xhigh): Accuracy 59.7%, mean time per trial 2,209 s GPT-6 Astra (medium): Accuracy 55.1%, mean time per trial 1,886 s GPT-6 Astra (low): Accuracy 50.9%, mean time per trial 1,684 s Grok 4.7 (xhigh): Accuracy 38.8%, mean time per trial 5,701 s MiMo-V2.6 Flash (none): Accuracy 16.1%, mean time per trial 161 s MiMo-V2.6 Pro (none): Accuracy 27%, mean time per trial 168 s GPT-6 Luna (max): Accuracy 15.2%, mean time per trial 2,284 s GPT-6 Sol (max): Accuracy 50.3%, mean time per trial 2,662 s Claude Sonnet 5.5 (max): Accuracy 63.9%, mean time per trial 7,432 s Beam 1.0 (none): Accuracy 6.36%, mean time per trial 179 s Mistral Large 4 (max): Accuracy 27.6%, mean time per trial 143 s Claude Haiku 5.5 (max): Accuracy 40%, mean time per trial 141 s Nemotron 3.5 Super VL (none): Accuracy 0.3%, mean time per trial 136 s DeepSeek V4.1 Flash 1017 (max): Accuracy 38.5%, mean time per trial 148 s Tencent Hy4 (max): Accuracy 37.9%, mean time per trial 126 s Inkling 2 Small (xhigh): Accuracy 13%, mean time per trial 7,273 s Claude Fable 5.2 (max): Accuracy 65.8%, mean time per trial 7,333 s Muse Ember (xhigh): Accuracy 30.6%, mean time per trial 2,288 s Gemini 4 Hydrogen (high): Accuracy 56.4%, mean time per trial 1,926 s GLM-5.3-Flash (max): Accuracy 37%, mean time per trial 120 s GPT-6 Astra (max): Accuracy 60.3%, mean time per trial 2,796 s GPT-6 Astra (high): Accuracy 60%, mean time per trial 2,114 s Claude Opus 5.5 (max): Accuracy 65.8%, mean time per trial 4,790 s GPT-6.1 Sol (max): Accuracy 60.9%, mean time per trial 3,177 s Kimi K3.1 Pro (max): Accuracy 59.7%, mean time per trial 123 s GPT-6.1 Terra (max): Accuracy 53.3%, mean time per trial 123 s Claude Opus 5.5 (max)GPT-6.1 Sol (max)GPT-6 Astra (max)GPT-6 Astra (high)Kimi K3.1 Pro (max)GPT-6.1 Terra (max)GLM-5.3-Flash (max)
ModelSettingAccuracyMean time per trial
Claude Opus 5.5 max, Claude Code 65.8% 4,790 s
Claude Fable 5.2 max, Claude Code 65.8% 7,333 s
Claude Sonnet 5.5 max, Claude Code 63.9% 7,432 s
GPT-6.1 Sol max, Codex 60.9% 3,177 s
GPT-6 Astra max, Codex 60.3% 2,796 s
GPT-6 Astra high, Codex 60% 2,114 s
Kimi K3.1 Pro max, Claude Code 59.7% 123.2 s
GPT-6 Astra xhigh, Codex 59.7% 2,209 s
Claude Fable 5.1 xhigh, Claude Code 57.6% 3,202 s
Claude Fable 5.1 max, Claude Code 57.3% 3,894 s
Gemini 4 Hydrogen high, Antigravity CLI 56.4% 1,926 s
GPT-6 Astra medium, Codex 55.2% 1,886 s
Claude Fable 5.1 high, Claude Code 54.9% 3,018 s
Claude Fable 5.1 medium, Claude Code 53.9% 2,528 s
GPT-6.1 Terra max, Codex 53.3% 123.1 s
Claude Opus 5 xhigh, Claude Code 53.3% 4,515 s
Claude Opus 5 max, Claude Code 52.1% 4,793 s
GPT-6 Astra low, Codex 50.9% 1,684 s
GPT-6 Sol max, Codex 50.3% 2,662 s
Claude Opus 5 high, Claude Code 49.7% 3,822 s
Claude Opus 5 medium, Claude Code 45.2% 3,293 s
Claude Fable 5 max, Claude Code 44.9% 4,203 s
Claude Fable 5.1 low, Claude Code 43.3% 2,291 s
GLM-5.3 max, Claude Code 41.2% 5,831 s
Claude Haiku 5.5 max, Claude Code 40% 140.5 s
Grok 4.7 xhigh, Grok Build 38.8% 5,701 s
DeepSeek V4.1 Flash 1017 max, Claude Code 38.5% 148.4 s
GPT-5.6 Sol max, Codex 38.5% 2,388 s
Tencent Hy4 max, Claude Code 37.9% 126.2 s
GLM-5.3-Flash max, Claude Code 37% 120.4 s
GLM-5.3-Flash none, Claude Code 35.5% 6,210 s
Claude Opus 5 low, Claude Code 34.9% 2,947 s
Muse Ember xhigh, Muse Code 30.6% 2,288 s
Mistral Large 4 max, Claude Code 27.6% 142.7 s
MiMo-V2.6 Pro none, mini-swe-agent 27% 168.2 s
Qwen3.8 Max (0902) max, Claude Code 27% 9,953 s
Claude Opus 4.8 max, Claude Code 23.6% 5,163 s
Kimi K3 max, Claude Code 23% 141.9 s
GPT-5.6 Terra max, Codex 21.5% 2,515 s
Grok 4.6 high, Grok Build 19.7% 2,352 s
Gemini 3.8 Flash high, mini-SWE-agent 18.2% 1,995 s
GPT-5.6 Luna max, Codex 17.6% 4,088 s
MiMo-V2.6 Flash none, mini-swe-agent 16.1% 160.8 s
GPT-6 Luna max, Codex 15.2% 2,284 s
Muse Spark 1.3 xhigh, Muse Code 14.9% 4,342 s
Inkling 2 Small xhigh, mini-swe-agent 13% 7,273 s
Gemini 3.7 Flash high, mini-SWE-agent 12.4% 1,689 s
Grok 4.5 high, Grok Build 12.4% 3,162 s
Claude Sonnet 5 max, Claude Code 12.1% 6,510 s
MiniMax-M3 none, mini-swe-agent 7.6% 204 s
DeepSeek V4 Pro 0813 max, Claude Code 6.7% 138.2 s
Beam 1.0 none, mini-swe-agent 6.4% 178.6 s
GPT-5.4 Mini xhigh, Codex 1.8% 141.5 s
Inkling-Small none, mini-swe-agent 0.9% 141.3 s
Nemotron 3.5 Super VL none, mini-swe-agent 0.3% 135.8 s
Inkling none, mini-swe-agent 0% 174.1 s

The best score, 65.8%, costs $14 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at max comes within 4.9 points of the best for $1.92 a task, 7 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Thinking Machines Z.ai Anthropic OpenAI DeepSeek Cost frontier One model at each effort level
0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5, max GPT-6.1 Sol, max DeepSeek V4.1 Flash 1017, max GPT-6 Luna, max Inkling-Small, none 0% 20% 40% 60% 80% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5, max GPT-6.1 Sol, max

Source: Official Terminal-Bench 4.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 max Claude Code 65.8% $14 269,135
Claude Fable 5.2 max Claude Code 65.8% $50 491,352
Claude Sonnet 5.5 max Claude Code 63.9% $22 565,388
GPT-6.1 Sol max Codex 60.9% $1.92 94,892
GPT-6 Astra max Codex 60.3% $9.90 72,694
GPT-6 Astra high Codex 60% $6.88 41,527
GPT-6 Astra xhigh Codex 59.7% $7.12 45,304
Kimi K3.1 Pro max Claude Code 59.7% $3.29 80,956
Claude Fable 5.1 xhigh Claude Code 57.6% $15 152,809
Claude Fable 5.1 max Claude Code 57.3% $19 191,277
Gemini 4 Hydrogen high Antigravity CLI 56.4% $2.38 119,482
GPT-6 Astra medium Codex 55.1% $5.80 32,638
Claude Fable 5.1 high Claude Code 54.9% $12 113,421
Claude Fable 5.1 medium Claude Code 53.9% $8.59 78,763
Claude Opus 5 xhigh Claude Code 53.3% $18 179,281
GPT-6.1 Terra max Codex 53.3% $2.65 100,138
Claude Opus 5 max Claude Code 52.1% $18 200,037
GPT-6 Astra low Codex 50.9% $4.72 24,207
GPT-6 Sol max Codex 50.3% $4.40 92,642
Claude Opus 5 high Claude Code 49.7% $14 144,242
Claude Opus 5 medium Claude Code 45.1% $9.67 101,450
Claude Fable 5 max Claude Code 44.9% $22 177,636
Claude Fable 5.1 low Claude Code 43.3% $7.15 60,409
GLM-5.3 max Claude Code 41.2% $8.27 208,061
Claude Haiku 5.5 max Claude Code 40% $2.89 386,483
Grok 4.7 xhigh Grok Build 38.8% $11 69,627
GPT-5.6 Sol max Codex 38.5% $7.70 71,065
DeepSeek V4.1 Flash 1017 max Claude Code 38.5% $0.56 497,265
Tencent Hy4 max Claude Code 37.9% $1.05 368,811
GLM-5.3-Flash max Claude Code 37% $1.14 459,702
GLM-5.3-Flash none Claude Code 35.5% $0.36 135,014
Claude Opus 5 low Claude Code 34.9% $7.25 72,229
Muse Ember xhigh Muse Code 30.6% $0.48 165,351
Mistral Large 4 max Claude Code 27.6% $2.18 253,650
Qwen3.8 Max (0902) max Claude Code 27% $14 252,708
MiMo-V2.6 Pro none mini-swe-agent 27% $0.86 245,849
Claude Opus 4.8 max Claude Code 23.6% $20 269,131
Kimi K3 max Claude Code 23% $5.82 195,427
GPT-5.6 Terra max Codex 21.5% $5.25 100,681
Grok 4.6 high Grok Build 19.7% $11 98,843
Gemini 3.8 Flash high mini-SWE-agent 18.2% $5.54 206,232
GPT-5.6 Luna max Codex 17.6% $1.05 185,705
MiMo-V2.6 Flash none mini-swe-agent 16.1% $0.59 200,090
GPT-6 Luna max Codex 15.2% $0.22 137,851
Muse Spark 1.3 xhigh Muse Code 14.8% 265,112
Inkling 2 Small xhigh mini-swe-agent 13% $2.92 309,788
Grok 4.5 high Grok Build 12.4% $6.35 98,149
Gemini 3.7 Flash high mini-SWE-agent 12.4% $3.82 157,982
Claude Sonnet 5 max Claude Code 12.1% $29 353,330
MiniMax-M3 none mini-swe-agent 7.58% $3.48 234,421
DeepSeek V4 Pro 0813 max Claude Code 6.67% $2.20 245,205
Beam 1.0 none mini-swe-agent 6.36% $0.67 64,523
GPT-5.4 Mini xhigh Codex 1.82% $2.04 169,597
Inkling-Small none mini-swe-agent 0.91% $0.15 20,104
Nemotron 3.5 Super VL none mini-swe-agent 0.30% $0.51 159,709
Inkling none mini-swe-agent 0% $3.06 39,308

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Own leaderboard 1 leader
Claude Opus 5.5 65.8%

Source: Official Terminal-Bench 4.1 leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Opus 5.5 GPT-6 Astra Claude Opus 5 Claude Fable 5 Claude Opus 4.8 GPT-5.4 Mini 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Claude Opus 5.5 Claude Fable 5.1

Source: Official Terminal-Bench 4.1 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Google DeepMind Anthropic OpenAI
Terminal-Bench 4.1 20 leaders
Claude Opus 5.5 65.8%
Terminal-Bench Replaced by 2.0, Nov 2025
GPT-5 49.6%
Terminal-Bench 2.0 Replaced by 2.1, May 2026
GPT-5.5 84.7%
Terminal-Bench 2.1 Passed 85% in Sep 2026, by GPT-6 Astra
GPT-6 Astra 87.4%
Terminal-Bench 3.0 Replaced by 4.0, Aug 2026
Claude Opus 5 42.7%
Terminal-Bench 4.0 4 leaders
Claude Opus 5.5 64.8%
Terminal-Bench 4.1 1 leader
Claude Opus 5.5 65.8%

Sources: Official leaderboards and boards copied by Epoch AI (CC BY 4.0)

Replaces Terminal-Bench 4.0

More in Terminal agents

All 19 in Terminal agents
  • SkillsBench In use 86-87 containerized expert tasks in 11 domains, run with and without Agent Skills. Verifier tests give pass rate and skill lift. GPT-5.5 67.3%
  • Terminal-Bench Hard In use Agents complete 44 hard tasks from Terminal-Bench Core in the Terminus 2 harness. Artificial Analysis scores pass@1 averaged over 3 repeats. Gemma 4 31B IT 36%
  • Harbor-Index In use Agents solve about 80 hard tasks drawn from 29 benchmarks run through Harbor adapters, such as HLE, GSO and SWE-bench Pro. Each task keeps its source verifier. GPT-5.5 28.1%
  • Terminal-Bench 2.1 Extended Not in model cards Agents complete long command-line tasks in containers, scored by verification tests. Mercor adds 99 new expert-written tasks to the 89 public ones and reports both scores. Muse Wonder 54%
  • o11y-bench Not in model cards Agents query metrics, logs and traces in a live Grafana stack, investigate incidents and fix dashboards. Grading checks claims against facts read from the same environment. Claude Opus 5.5 93.8%
  • CompileBench Not in model cards Agents build real open-source projects through dependency problems, legacy toolchains and complex build systems, for example compiling 22-year-old code. Claude Opus 4.1 100%