Skip to content

Terminal-Bench 2.1

Terminal-Bench 2.1 results, mostly provider-reported, stored separately from the Terminal-Bench 2.0 lane.

68 models

Higher is better.

  1. 1SWE-2Cognition92.8
  2. 2GPT-5.6 SolOpenAI91.9
  3. 3DeepSeek V4.1 Flash 1017DeepSeek90.6
  4. 4MiMo-V2.6-ProXiaomi89.9
  5. 5Gemini 3.8 FlashGoogle89.4
  6. 6Muse Spark 1.3Meta88.8
  7. 7Kimi K3Moonshot AI88.3
  8. 8GLM-5.3Z.AI88.2
  9. 9Claude Mythos 5Anthropic88.0
  10. 10DeepSeek V4 Pro 0813DeepSeek87.9
  11. 11MiMo-V2.6-FlashXiaomi87.6
  12. 12GPT-5.6 TerraOpenAI87.4
  13. 13Qwen3.8 MaxAlibaba86.6
  14. 14Ornith-1.5-397BOrnith AI86.1
  15. 15Gemini 3.7 FlashGoogle85.8
  16. 16Hy4 previewTencent85.4
  17. 17Step 5 PreviewStepFun85.0
  18. 18GPT-5.6 LunaOpenAI84.7
  19. 19Claude Fable 5Anthropic84.3
  20. 20GLM-5.3-FlashZ.AI84.3
  21. 21Grok 4.5xAI83.3
  22. 22Ultramarine 1 FlashPaper Instruments83.2
  23. 23Muse Spark 1.2Meta82.9
  24. 24DeepSeek V4 Flash 0731DeepSeek82.7
  25. 25Ember-1Fireworks82.0
  26. 26SWE-1.7Cognition81.5
  27. 27Ling 3.1 FlashInclusionAI81.2
  28. 28GLM-5.2Z.AI81.0
  29. 29Claude Sonnet 5Anthropic80.4
  30. 30Beam 1.0 501B-A23BReflection AI80.1
  31. 31Muse Spark 1.1Meta80.0
  32. 32Atria Dawn PreviewShanghai Artificial Intelligence Laboratory78.3
  33. 33Ornith-1.0-397BDeepReinforce AI77.5
  34. 34Gemini 3.5 FlashGoogle76.2
  35. 35dots3-note PreviewDots Studio75.1
  36. 36Claude Opus 4.8Anthropic74.6
  37. 37Jais-3-375B-A23BIFM73.8
  38. 38Qwen3.8-27BAlibaba73.0
  39. 39Seed 2.1 ProByteDance71.0
  40. 40Apodex 1.1Apodex70.8
Higher is better. Showing the top 40 of 68. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Interactive task success rate
Tasks
Terminal-based software-agent tasks
Authors
Terminal-Bench contributors (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.

All results

1SWE-2 Cognition92.8
2GPT-5.6 Sol OpenAI91.9
3DeepSeek V4.1 Flash 1017 DeepSeek90.6
4MiMo-V2.6-Pro Xiaomi89.9
5Gemini 3.8 Flash Google89.4
6Muse Spark 1.3 Meta88.8
7Kimi K3 Moonshot AI88.3
8GLM-5.3 Z.AI88.2
9Claude Mythos 5 Anthropic88.0
10DeepSeek V4 Pro 0813 DeepSeek87.9
11MiMo-V2.6-Flash Xiaomi87.6
12GPT-5.6 Terra OpenAI87.4
13Qwen3.8 Max Alibaba86.6
14Ornith-1.5-397B Ornith AI86.1
15Gemini 3.7 Flash Google85.8
16Hy4 preview Tencent85.4
17Step 5 Preview StepFun85.0
18GPT-5.6 Luna OpenAI84.7
19Claude Fable 5 Anthropic84.3
20GLM-5.3-Flash Z.AI84.3
21Grok 4.5 xAI83.3
22Ultramarine 1 Flash Paper Instruments83.2
23Muse Spark 1.2 Meta82.9
24DeepSeek V4 Flash 0731 DeepSeek82.7
25Ember-1 Fireworks82.0
26SWE-1.7 Cognition81.5
27Ling 3.1 Flash InclusionAI81.2
28GLM-5.2 Z.AI81.0
29Claude Sonnet 5 Anthropic80.4
30Beam 1.0 501B-A23B Reflection AI80.1
31Muse Spark 1.1 Meta80.0
32Atria Dawn Preview Shanghai Artificial Intelligence Laboratory78.3
33Ornith-1.0-397B DeepReinforce AI77.5
34Gemini 3.5 Flash Google76.2
35dots3-note Preview Dots Studio75.1
36Claude Opus 4.8 Anthropic74.6
37Jais-3-375B-A23B IFM73.8
38Qwen3.8-27B Alibaba73.0
39Seed 2.1 Pro ByteDance71.0
40Apodex 1.1 Apodex70.8
41Laguna S 2.1 Poolside70.2
42Quasar 438B Multiverse Computing69.3
43Ornith-1.5-35B-A3B Ornith AI67.8
44Jais-3-MoVA-36B-A4B IFM67.3
45Nemotron 3.5 Super VL 120B A12B NVIDIA66.7
46MiniMax M3 MiniMax66.0
47Pokee-Isaac 28B Pokee AI65.1
48Inkling-Small Thinking Machines Lab64.7
49Inkling Thinking Machines Lab63.8
50Jais-3-32B IFM63.7
51MAI-Code-1.1-Flash Microsoft62.9
52Step 3.7 Flash StepFun59.5
53Ling 3.0 Flash InclusionAI57.0
54Solar Pro 4 Upstage57.0
55Nemotron 3 Ultra NVIDIA56.4
56Gemini 3.5 Flash-Lite Google54.0
57Ternary Bonsai 2 27B Prism ML52.8
58Muse Glimmer 30B Meta51.7
59Nemotron 3.5 Bolt 74B-A9B NVIDIA51.0
60Jais-3-7B IFM46.4
61Ornith-1.5-9B Ornith AI46.2
62K-EXAONE 2.0 LG AI Research43.8
63Jais-3-3.7B IFM41.2
64A.X K2 SK Telecom36.0
65Nemotron 3.5 Lightning 30B A3B NVFP4 NVIDIA23.5
66Granite 4.2 8B IBM20.6
67Mellum2.1-12B-A2.5B-Thinking JetBrains17.4
68MiniCPM5-2B OpenBMB8.6