terminal-bench/
Run by mupt-ai · Released 2026-11-09 · 87 tasks · 5 settings
Model Tradeoffs
Which coding agent works best on terminal-bench/terminal-bench-2-1? Most accurate: Dari Router, 83.9% at $0.11 per task. Best for less: GPT-6 Luna, 65.5% at $0.018. 5 model settings scored on 87 tasks from the benchmark.
All Settings
| Model | Harness | Reasoning | Access | Accuracy | Cost / Task | Frontier |
|---|---|---|---|---|---|---|
| Dari Router | Pi | Default | API Key | 83.9% | $0.11 | Yes |
| accounts/fireworks/routers/kimi-fast-latest | Pi | Default | Custom Endpoint | 82.8% | $0.51 | No |
| switchyard | Codex | Default | Custom Endpoint | 79.3% | $0.14 | No |
| GPT-6 Luna | Codex | Low | API Key | 65.5% | $0.018 | Yes |
| GPT-6 Luna | Codex | High | API Key | 64.4% | $0.011 | Yes |