Terminal-Bench
Agentic terminal tasks
2.1agents and tool useactive
Score in %
Definition
- Version
- 2.1
- Category
- agents and tool use
- Maintainer
- Terminal-Bench
- Status
- active
- Home page
- https://www.tbench.ai/leaderboard
Sources
- Claude Fable 5.1 Benchmarks, Cost and Capabilities | Vals AI vals.ai https://vals.ai/models/anthropic_claude-fable-5-1
- huggingface.co/moonshotai/Kimi-K3 https://huggingface.co/moonshotai/Kimi-K3
Scores at a Glance
Score in %
Higher is better for this unit.
1
SWE-2Cognition (owns Windsurf)
DeepSeek-V4.1-FlashDeepSeek
MiMo-V2.6-ProXiaomi
Kimi K3Moonshot AI
GLM-5.3Z.ai (Zhipu)
Qwen3.8-2.4T-A95BAlibaba (Qwen)
Hy4-previewTencent
Claude Fable 5.1AnthropicVals.ai, max effort; 79.03% if fallback-assisted tasks count as failuresGLM-5.2Z.ai (Zhipu)Qwen3.8-27BAlibaba (Qwen)Terminus
Mellum2.1-12B-A2.5B-ThinkingJetBrainsthinking mode
SWE-2Cognition (owns Windsurf)92.8 %
vendor-reported2DeepSeek-V4.1-FlashDeepSeek90.6 %
vendor-reported3MiMo-V2.6-ProXiaomi89.9 %
vendor-reported4Kimi K3Moonshot AI88.3 %
vendor-reported5GLM-5.3Z.ai (Zhipu)88.2 %
vendor-reported6Qwen3.8-2.4T-A95BAlibaba (Qwen)86.6 %
vendor-reported7Hy4-previewTencent85.4 %
vendor-reported8Claude Fable 5.1AnthropicVals.ai, max effort; 79.03% if fallback-assisted tasks count as failures85.02 %
independent9GLM-5.2Z.ai (Zhipu)81 %
vendor-reported10Qwen3.8-27BAlibaba (Qwen)Terminus73 %
vendor-reported11Mellum2.1-12B-A2.5B-ThinkingJetBrainsthinking mode17.4 %
vendor-reported0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
| SWE-2 | 92.8 % | vendor-reported | – | – | – | orcarouter.ai |
| DeepSeek-V4.1-Flash | 90.6 % | vendor-reported | – | – | – | huggingface.co |
| DeepSeek-V4.1-Flash | 90.6 % | vendor-reported | DeepSeek Harness Minimal | – | – | huggingface.co |
| MiMo-V2.6-Pro | 89.9 % | vendor-reported | – | – | – | huggingface.co |
| Kimi K3 | 88.3 % | vendor-reported | – | – | – | huggingface.co |
| Kimi K3 | 88.3 % | vendor-reported | Kimi Code harness | – | – | huggingface.co |
| GLM-5.3 | 88.2 % | vendor-reported | – | – | – | huggingface.co |
| GLM-5.3 | 88.2 % | vendor-reported | Claude Code 2.1.207 | – | – | huggingface.co |
| MiMo-V2.6-Flash | 87.6 % | vendor-reported | – | MiMo-V2.6 model card comparison table | 21 Sep 2026 | huggingface.co |
| Qwen3.8-2.4T-A95B | 86.6 % | vendor-reported | – | – | – | huggingface.co |
| Hy4-preview | 85.4 % | vendor-reported | – | – | – | huggingface.co |
| Claude Fable 5.1 | 85.02 % | independent | Vals.ai, max effort; 79.03% if fallback-assisted tasks count as failures | – | – | vals.ai |
| GLM-5.2 | 81 % | vendor-reported | – | – | – | rits.shanghai.nyu.edu |
 Motif-3 | 74.9 % | vendor-reported | temperature 1.0, top-p 0.95, max sequence 262,144 | Motif-3 model card | Aug 2026 | huggingface.co |
| Qwen3.8-27B | 73 % | vendor-reported | Terminus | – | – | huggingface.co |
 A.X K2 | 36 % | vendor-reported | Thinking mode | A.X K2 model card | Jul 2026 | huggingface.co |
| Mellum2.1-12B-A2.5B-Thinking | 17.4 % | vendor-reported | thinking mode | – | – | huggingface.co |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Terminal-Bench", https://aurion.semai.id/benchmarks/terminal-bench-2-1/, records of October 2026.