Benchmarks / Agentic
Long-Horizon Terminal-Bench
An agent works in a terminal for up to 90 minutes per task on 46 long jobs — building systems, migrating frameworks, playing games move by move — and earns partial credit for real progress. Maintained by Long-Horizon Terminal-Bench authors. Score: Mean reward over the 46 tasks (hidden verifiers pay continuous partial credit from 0 to 1; errors count 0).
Long-Horizon Terminal-Bench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Grok 4.5 | max | 50.5% |
| 2 | Sonnet 5 | max | 49.7% |
| 3 | Opus 4.8 | max | 49.2% |
| 4 | Fable 5 | max | 48.7% |
| 5 | GPT-5.6 Sol | max | 45.1% |
| 6 | GPT-5.5 | xhigh | 44.5% |
| 7 | MiniMax M3 | max | 38.5% |
| 8 | Kimi K3 | max | 37.8% |
| 9 | Sonnet 4.6 | max | 37.3% |
| 10 | Kimi K2.7 Code | max | 36.7% |
| 11 | GLM 5.2 | max | 31.6% |
| 12 | Qwen 3.6 Plus | max | 31.3% |
| 13 | DeepSeek V4 Pro | max | 30.7% |
| 14 | Qwen 3.7 Max | max | 29.6% |
| 15 | Hy3 | xhigh | 28.8% |
| 16 | Gemini 3.1 Pro | high | 27.9% |
| 17 | GPT-5.4 | xhigh | 27.2% |
| 18 | GLM 5.1 | max | 26.7% |
| 19 | Kimi K2.6 | max | 25.5% |
| 20 | GPT-5.3 Codex | xhigh | 21.5% |
| 21 | Grok 4.20 | xhigh | 10.2% |