Benchmark Almanac

Terminal-Bench 2.1 Extended

Muse Wonder 54% leads since October 2026
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.

Held-out extension of terminal-bench-2-1. Mercor shows the original and extended score for each model. The month is approximate. The Wayback Machine first saved the earliest extended pages on August 15th, 2026.

Made by
Mercor
Released
August 2026
Kind
Terminal tasks
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Strongest model tested
Claude Opus 5.5 #4 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
46
Entries on its own leaderboard, 44 models we track

Muse Wonder leads with 54%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

46 entries. Scroll sideways for the rest.

  1. 54.0 Muse Wonder
  2. 47.5 Claude Opus 5.5
  3. 46.0 Muse Spark Open
  4. 45.8 GPT-6.1 Sol
  5. 45.5 Grok 4.6
  6. 45.5 Claude Sonnet 5.5
  7. 44.4 Gemini 3.7 Flash
  8. 44.4 Gemini 3.8 Flash
  9. 42.4 GPT-6 Sol
  10. 40.1 Gemini 3.6 Flash
  11. 39.4 GLM-5.3
  12. 38.4 DeepSeek V4.1 Flash
  13. 38.0 GPT-5.5
  14. 38.0 Claude Opus 5
  15. 37.7 Claude Haiku 5.5
  16. 37.4 GPT-5.6 Sol
  17. 36.4 GPT-5.6 Terra
  18. 35.7 Beam
  19. 35.4 GPT-5.6 Luna
  20. 35.4 Qwen 3.8 Max
  21. 35.0 Gemini 3.1 Pro
  22. 33.3 Grok 4.5
  23. 33.3 GPT-6 Luna
  24. 32.7 Claude Opus 4.8
  25. 31.3 GPT-5.4
  26. 31.0 Kimi K3
  27. 30.0 DeepSeek-V4-Flash
  28. 30.0 DeepSeek V4 Pro 0813
  29. 30.0 GLM-5.3-Flash
  30. 28.9 Claude Sonnet 5
  31. 28.6 Gemini 3.5 Flash
  32. 27.9 GLM-5.2
  33. 27.6 Claude Opus 4.7
  34. 24.6 Inkling
  35. 23.6 Kimi K2.7 Code
  36. 22.2 MiniMax-M3
  37. 21.2 DeepSeek-V4-Pro
  38. 19.5 Claude Fable 5.1
  39. 17.8 Qwen3.5
  40. 17.5 Claude Sonnet 4.6
  41. 16.8 Claude Fable 5
  42. 15.2 DeepSeek-V3.2
  43. 11.8 Nemotron 3 Ultra
  44. 11.8 Grok 4.7
  45. 11.1 Kimi K2
  46. 10.1 MiniMax-M2.7
  1. 1 Muse Wonder max 54%
  2. 2 Claude Opus 5.5 max 47.5%
  3. 3 Muse Spark Open xhigh 46%
  4. 4 GPT-6.1 Sol max 45.8%
  5. 5 Grok 4.6 xhigh 45.5%
  6. 6 Claude Sonnet 5.5 max 45.5%
  7. 7 Gemini 3.7 Flash high 44.4%
  8. 8 Gemini 3.8 Flash high 44.4%
  9. 9 GPT-6 Sol max 42.4%
  10. 10 Gemini 3.6 Flash high 40.1%
  11. 11 GLM-5.3 max 39.4%
  12. 12 DeepSeek V4.1 Flash max 38.4%
  13. 13 GPT-5.5 xhigh 38%
  14. 14 Claude Opus 5 max 38%
  15. 15 Claude Haiku 5.5 max 37.7%
  16. 16 GPT-5.6 Sol xhigh 37.4%
  17. 17 GPT-5.6 Terra max 36.4%
  18. 18 Beam max 35.7%
  19. 19 GPT-5.6 Luna max 35.4%
  20. 20 Qwen 3.8 Max xhigh 35.4%
  21. 21 Gemini 3.1 Pro high 35%
  22. 22 Grok 4.5 high 33.3%
  23. 23 GPT-6 Luna max 33.3%
  24. 24 Claude Opus 4.8 max 32.7%
  25. 25 GPT-5.4 xhigh 31.3%
  26. 26 Kimi K3 max 31%
  27. 27 DeepSeek-V4-Flash max 30%
  28. 28 DeepSeek V4 Pro 0813 max 30%
  29. 29 GLM-5.3-Flash max 30%
  30. 30 Claude Sonnet 5 high 28.9%
  31. 31 Gemini 3.5 Flash high 28.6%
  32. 32 GLM-5.2 max 27.9%
  33. 33 Claude Opus 4.7 high 27.6%
  34. 34 Inkling high 24.6%
  35. 35 Kimi K2.7 Code high 23.6%
  36. 36 MiniMax-M3 high 22.2%
  37. 37 DeepSeek-V4-Pro max 21.2%
  38. 38 Claude Fable 5.1 max 19.5%
  39. 39 Qwen3.5 17.8%
  40. 40 Claude Sonnet 4.6 high 17.5%
  41. 41 Claude Fable 5 max 16.8%
  42. 42 DeepSeek-V3.2 15.2%
  43. 43 Nemotron 3 Ultra high 11.8%
  44. 44 Grok 4.7 xhigh 11.8%
  45. 45 Kimi K2 Sep 2025, high 11.1%
  46. 46 MiniMax-M2.7 high 10.1%

Source: Official Terminal-Bench 2.1 Extended leaderboard

#EntryScoreSettingDateSource
1 Muse Wonder 54% max Oct 2026 Official leaderboard
2 Claude Opus 5.5 47.5% max Sep 2026 Official leaderboard
3 Muse Spark Open 46% xhigh Oct 2026 Official leaderboard
4 GPT-6.1 Sol 45.8% max Sep 2026 Official leaderboard
5 Grok 4.6 45.5% xhigh Aug 2026 Official leaderboard
6 Claude Sonnet 5.5 45.5% max Sep 2026 Official leaderboard
7 Gemini 3.7 Flash 44.4% high Aug 2026 Official leaderboard
8 Gemini 3.8 Flash 44.4% high Sep 2026 Official leaderboard
9 GPT-6 Sol 42.4% max Sep 2026 Official leaderboard
10 Gemini 3.6 Flash 40.1% high Jul 2026 Official leaderboard
11 GLM-5.3 39.4% max Aug 2026 Official leaderboard
12 DeepSeek V4.1 Flash 38.4% max Sep 2026 Official leaderboard
13 GPT-5.5 38% xhigh Apr 2026 Official leaderboard
14 Claude Opus 5 38% max Jul 2026 Official leaderboard
15 Claude Haiku 5.5 37.7% max Oct 2026 Official leaderboard
16 GPT-5.6 Sol 37.4% xhigh Jul 2026 Official leaderboard
17 GPT-5.6 Terra 36.4% max Jul 2026 Official leaderboard
18 Beam 35.7% max Official leaderboard
19 GPT-5.6 Luna 35.4% max Jul 2026 Official leaderboard
20 Qwen 3.8 Max 35.4% xhigh Aug 2026 Official leaderboard
21 Gemini 3.1 Pro 35% high Feb 2026 Official leaderboard
22 Grok 4.5 33.3% high Jul 2026 Official leaderboard
23 GPT-6 Luna 33.3% max Sep 2026 Official leaderboard
24 Claude Opus 4.8 32.7% max May 2026 Official leaderboard
25 GPT-5.4 31.3% xhigh Mar 2026 Official leaderboard
26 Kimi K3 31% max Jul 2026 Official leaderboard
27 DeepSeek-V4-Flash 30% max Apr 2026 Official leaderboard
28 DeepSeek V4 Pro 0813 30% max Aug 2026 Official leaderboard
29 GLM-5.3-Flash 30% max Aug 2026 Official leaderboard
30 Claude Sonnet 5 28.9% high Jun 2026 Official leaderboard
31 Gemini 3.5 Flash 28.6% high May 2026 Official leaderboard
32 GLM-5.2 27.9% max Jun 2026 Official leaderboard
33 Claude Opus 4.7 27.6% high Apr 2026 Official leaderboard
34 Inkling 24.6% high Jul 2026 Official leaderboard
35 Kimi K2.7 Code 23.6% high Jun 2026 Official leaderboard
36 MiniMax-M3 22.2% high Jun 2026 Official leaderboard
37 DeepSeek-V4-Pro 21.2% max Apr 2026 Official leaderboard
38 Claude Fable 5.1 19.5% max Sep 2026 Official leaderboard
39 Qwen3.5 17.8% Official leaderboard
40 Claude Sonnet 4.6 17.5% high Feb 2026 Official leaderboard
41 Claude Fable 5 16.8% max Jun 2026 Official leaderboard
42 DeepSeek-V3.2 15.2% Dec 2025 Official leaderboard
43 Nemotron 3 Ultra 11.8% high Jun 2026 Official leaderboard
44 Grok 4.7 11.8% xhigh Sep 2026 Official leaderboard
45 Kimi K2 Sep 2025 11.1% high Sep 2025 Official leaderboard
46 MiniMax-M2.7 10.1% high Mar 2026 Official leaderboard

More effort adds a median 3 points

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude Opus 5
high max
38%
Claude Opus 4.8
high max
32.7%
Claude Opus 4.7
high
27.6%

Source: Official Terminal-Bench 2.1 Extended leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Muse Wonder max 54%
Claude Opus 5.5 max 47.5%
Muse Spark Open xhigh 46%
GPT-6.1 Sol max 45.8%
Grok 4.6 xhigh 45.5%
Claude Sonnet 5.5 max 45.5%
Gemini 3.7 Flash high 44.4%
Gemini 3.8 Flash high 44.4%
GPT-6 Sol max 42.4%
Gemini 3.6 Flash high 40.1%
GLM-5.3 max 39.4%
DeepSeek V4.1 Flash max 38.4%
GPT-5.5 xhigh 38%
Claude Opus 5 max 38%
Claude Haiku 5.5 max 37.7%
GPT-5.6 Sol xhigh 37.4%
GPT-5.6 Terra max 36.4%
GPT-5.6 Luna max 35.4%
Qwen 3.8 Max xhigh 35.4%
Gemini 3.1 Pro high 35%
Claude Opus 5 high 34%
Grok 4.5 high 33.3%
GPT-6 Luna max 33.3%
Claude Opus 4.8 max 32.7%
GPT-5.4 xhigh 31.3%
Kimi K3 max 31%
DeepSeek-V4-Flash max 30%
DeepSeek V4 Pro 0813 max 30%
GLM-5.3-Flash max 30%
Claude Opus 4.8 high 29.3%
Claude Sonnet 5 high 28.9%
Gemini 3.5 Flash high 28.6%
GLM-5.2 max 27.9%
Claude Opus 4.7 high 27.6%
Claude Opus 4.7 max 27.3%
Inkling high 24.6%
Kimi K2.7 Code high 23.6%
MiniMax-M3 high 22.2%
DeepSeek-V4-Pro max 21.2%
Claude Fable 5.1 max 19.5%
Claude Sonnet 4.6 high 17.5%
Claude Fable 5 max 16.8%
DeepSeek-V3.2 15.2%
Nemotron 3 Ultra high 11.8%
Grok 4.7 xhigh 11.8%
Kimi K2 Sep 2025 high 11.1%
MiniMax-M2.7 high 10.1%

4 models have held number one on the official leaderboard since its release in August 2026

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

Google DeepMind xAI Anthropic Meta
Own leaderboard 4 leaders
Muse Wonder 54%

Source: Official Terminal-Bench 2.1 Extended 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 Near the ceiling (85%) Muse Wonder Claude Opus 5.5 Gemini 3.6 Flash GPT-5.5 Claude Sonnet 4.6 Kimi K2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Muse Wonder Claude Opus 5.5

Source: Official Terminal-Bench 2.1 Extended 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.

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 4.1 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. Claude Opus 5.5 65.8%
  • 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%