BenchmarksAgents
Terminal-Bench 2.1
A command-line agent benchmark for completing terminal tasks in reproducible task environments.
Agentstask successHigher is better
What is Terminal-Bench 2.1?
A command-line agent benchmark for completing terminal tasks in reproducible task environments. evals.report tracks reported Terminal-Bench 2.1 scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.
Top reported Terminal-Bench 2.1 score: GPT-5.6 Sol Ultra — 91.9% (task success).
| Model | Lab | Score↓ | Source model | Status | Date | |
|---|---|---|---|---|---|---|
| GPT-5.6 Sol Ultra | OpenAI | 91.9% | GPT-5.6 Sol (Ultra) | Verified | Jul 9, 2026 | Details |
| GPT-5.6 Sol | OpenAI | 88.8% | GPT-5.6 Sol | Verified | Jul 9, 2026 | Details |
| GPT-5.6 Terra | OpenAI | 87.4% | GPT-5.6 Terra | Verified | Jul 9, 2026 | Details |
| GPT-5.6 Luna | OpenAI | 84.7% | GPT-5.6 Luna | Verified | Jul 9, 2026 | Details |
| Codex CLI + GPT-5.5 | Agent systems | 83.4% | Codex CLI + GPT-5.5 | Verified | — | Details |
| Grok 4.5 | xAI | 83.3% | Grok 4.5 | Verified | Jul 8, 2026 | Details |
| Fugu Ultra | Sakana AI | 82.1% | Fugu Ultra | Verified | Jun 15, 2026 | Details |
| SWE-1.7 | Cognition | 81.5% | SWE-1.7 | Verified | Jul 8, 2026 | Details |
| GLM-5.2Open | Z.ai | 81.0% | GLM-5.2 | Verified | Jun 16, 2026 | Details |
| Claude Sonnet 5 | Anthropic | 80.4% | Claude Sonnet 5 | Verified | Jun 30, 2026 | Details |
| Fugu | Sakana AI | 80.2% | Fugu | Verified | Jun 15, 2026 | Details |
| Muse Spark 1.1 | Meta | 80.0% | Muse Spark 1.1 | Verified | Jul 9, 2026 | Details |
| Claude Code + Claude Opus 4.8 | Agent systems | 78.9% | Claude Code + Claude Opus 4.8 | Verified | — | Details |
| Terminus 2 + GPT-5.5 | Agent systems | 78.2% | Terminus 2 + GPT-5.5 | Verified | — | Details |
| Ornith-1.0-397BOpen | DeepReinforce | 77.5% | Ornith-1.0-397B (Terminus-2) | Verified | Jun 25, 2026 | Details |
| Nex-N2-ProOpen | Nex AGI | 75.3% | Nex-N2-Pro | Verified | Jun 2, 2026 | Details |
| Terminus 2 + Claude Opus 4.8 | Agent systems | 74.6% | Terminus 2 + Claude Opus 4.8 | Verified | — | Details |
| Terminus 2 + Gemini 3 Pro | Agent systems | 74.4% | Terminus 2 + Gemini 3 Pro | Verified | — | Details |
| Gemini CLI + Gemini 3.1 Pro | Agent systems | 70.7% | Gemini CLI + Gemini 3.1 Pro | Verified | — | Details |
| Terminus 2 + Gemini 3.1 Pro | Agent systems | 70.3% | Terminus 2 + Gemini 3.1 Pro | Verified | — | Details |
| Claude Code + Claude Opus 4.7 | Agent systems | 69.7% | Claude Code + Claude Opus 4.7 | Verified | — | Details |
| Gemini CLI + Gemini 3 Pro | Agent systems | 66.3% | Gemini CLI + Gemini 3 Pro | Verified | — | Details |
| Terminus 2 + Claude Opus 4.7 | Agent systems | 66.1% | Terminus 2 + Claude Opus 4.7 | Verified | — | Details |
| MiniMax M3Open | MiniMax | 66.0% | MiniMax M3 (Terminus 2) | Verified | Jun 1, 2026 | Details |
| InklingOpen | Thinking Machines Lab | 63.8% | Inkling | Verified | Jul 15, 2026 | Details |
| Nex-N2-miniOpen | Nex AGI | 60.7% | Nex-N2-mini | Verified | Jun 2, 2026 | Details |
| Step 3.7 FlashOpen | StepFun | 59.6% | Step 3.7 Flash | Verified | May 29, 2026 | Details |
| Claude Code + GLM 5.1 | Agent systems | 58.7% | Claude Code + GLM 5.1 | Verified | — | Details |
| NVIDIA Nemotron 3 UltraOpen | NVIDIA | 56.4% | Nemotron-3-Ultra-550B-A55B (BF16) | Verified | Jun 4, 2026 | Details |
Each row reports the model’s task success on Terminal-Bench 2.1. Click a row for the full run context.