Benchmark
Terminal-Bench 2.1
Terminal-Bench 2.1 is an updated release of the Terminal-Bench benchmark that tests AI agents' ability to operate a computer via the terminal. It…
- Modality
- text
- Categories
- reasoning, agents, code, tool_calling
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 28
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Claude Fable 5 | Anthropic | 0.843 | 2026-06-09 | evidence |
| DeepSeek-V4-Flash-0731 | DeepSeek | 0.827 | 2026-07-31 | evidence |
| DeepSeek-V4-Pro-0813 | DeepSeek | 0.879 | 2026-08-13 | evidence |
| Gemini 3.5 Flash-Lite | 0.54 | 2026-07-21 | evidence | |
| Gemini 3.6 Flash | 0.78 | 2026-07-21 | evidence | |
| Gemini 3.7 Flash | 0.858 | 2026-08-13 | evidence | |
| GLM-5.2 | Z.ai | 0.827 | 2026-06-16 | evidence |
| GLM-5.3 | Z.ai | 0.882 | 2026-08-14 | evidence |
| GPT-5.6 Luna | OpenAI | 0.847 | 2026-07-09 | evidence |
| GPT-5.6 Sol | OpenAI | 0.888 | 2026-07-09 | evidence |
| GPT-5.6 Terra | OpenAI | 0.874 | 2026-07-09 | evidence |
| Grok 4.5 | xAI | 0.833 | 2026-07-16 | evidence |
| Hy3 | Tencent | 0.717 | 2026-07-06 | evidence |
| Inkling-Small | Thinking Machines Lab | 0.647 | 2026-07-30 | evidence |
| Kimi K3 | Moonshot AI | 0.883 | 2026-07-16 | evidence |
| Laguna S 2.1 | Poolside | 0.702 | 2026-07-21 | evidence |
| MAI-Code-1.1-Flash | Microsoft | 0.629 | 2026-08-11 | evidence |
| MiniMax M3 | MiniMax | 0.66 | 2026-06-01 | evidence |
| Muse Glimmer-30B | Meta | 0.517 | 2026-08-10 | evidence |
| Muse Spark 1.1 | Meta | 0.8 | 2026-07-09 | evidence |
| Muse Spark 1.2 | Meta | 0.829 | 2026-08-05 | evidence |
| Nemotron 3 Ultra (550B A55B) | NVIDIA | 0.564 | 2026-06-04 | evidence |
| Nemotron 3.5 Lightning (30B A3B) | NVIDIA | 0.2458 | 2026-08-11 | evidence |
| Qwen3.8 Max | Qwen | 0.866 | 2026-08-02 | evidence |
| Qwen3.8-27B | Qwen | 0.73 | 2026-08-14 | evidence |
| Seed 2.1 Pro | ByteDance | 0.71 | 2026-06-24 | evidence |
| Seed 2.1 Turbo | ByteDance | 0.676 | 2026-06-24 | evidence |
| Solar Pro 4 | Upstage | 0.57 | 2026-08-06 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.