Terminal-Bench 2.0 - Medium
Agents · 2026-06-04
The Medium split of Terminal-Bench 2.0. This child benchmark separates a source-reported subtask or subtrack from the parent aggregate so scores at different grains do not share one benchmark_slug.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 79.3 |
| Opus 4.8 | 77.0 |
| Opus 4.7 | 72.7 |
| Gemini 3.1 Pro Preview | 72.7 |
| Gemini 3.5 Flash | 72.7 |
| Opus 4.6 | 70.9 |
| Opus 4.5 | 69.1 |
| Sonnet 4.6 | 69.1 |
| GPT-5.3-Codex | 69.1 |
| GPT-5.4 | 69.1 |
| Muse Spark | 69.1 |
| Qwen3.7-Max | 67.9 |
| DeepSeek-V4-Pro | 67.3 |
| Kimi K2.6 | 67.3 |
| Qwen3.6-Max-Preview | 67.3 |
| Gemini 3 Pro Preview | 63.6 |
| GLM-5.1 | 61.8 |
| GLM-5 | 61.8 |
| MiniMax M3 | 60.0 |
| MiniMax M2.7 | 56.4 |
| Qwen3.6 27B | 56.4 |
| GPT-5.4 Mini | 53.9 |
| Gemini 3 Flash Preview | 52.7 |
| GPT-5.1 | 52.7 |
| MiniMax M2.5 | 52.7 |
| Qwen3.5 Plus (2026-02-15) | 52.7 |
| Qwen3.6 Plus (2026-04-02) | 52.7 |
| Grok 4.3 | 52.1 |
| GPT-5.2 | 50.9 |
| Sonnet 4.5 | 49.1 |
| Grok 4.20 | 49.1 |
| Kimi K2.5 | 49.1 |
| MiniMax M2.1 | 49.1 |
| Haiku 4.5 | 47.3 |
| DeepSeek-V3.2 | 47.3 |
| GPT-5.4 Nano | 47.3 |
| Kimi K2 Thinking | 47.3 |
| Gemma 4 31B IT | 45.5 |
| GPT-5 | 45.5 |
| GLM-4.7 | 43.6 |