We just open sourced a tiny GPT-style cognitive core built in pure Rust.See our repository
Terminal-Bench 2.1
One harness to unlock the potential of all models
Compare Ante runs across models on the same Terminal-Bench 2.1 task set, using consistent parameters and verified benchmark results.
Best Accuracy82.7%
Leading Model
DeepSeek V4 Flash 0731max
Task Set89 tasks
Trials368 passed / 445 trials
Benchmark principlesRead more
We benchmark what we ship.Every eval uses a pinned public . No eval-only branches or benchmark-specific prompts.
The runs are auditable.Every result links its raw Harbor run, so anyone can inspect the trials behind the number.
The constraints are official.All trials follow the official Terminal-Bench parameters: 89 tasks, 5 trials per task, strict timeouts, and hardware limits.
Model org
All model orgsTB 2.1 · Same parameters: 89 tasks · 5 trials/task · Updated Aug 9, 2026
| # | Model | Same-model | Agent | Source | |||||
|---|---|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4 Flash 0731max | 82.7% ±1.79 SE | $68.41 | 38.9 min | #1 same-model | Ante0.preview.71 | Harbor link | Aug 9, 2026 | |
| 2 | Grok 4.5mediumⓘ | 80.9% ±1.27 SE | $242.57 | 8.4 min | #1 same-model | Ante0.preview.56 | PR #129 | Jul 10, 2026 | |
| 3 | GLM 5.2 | 74.6% ±2.06 SE | $260.11 | 11.3 min | #1 same-model | Ante0.preview.43 | Harbor link | Jun 20, 2026 | |
| 4 | DeepSeek V4 Pro | 69.1% ±2.24 SE | $26.34 | 48.4 min | #1 same-model | Ante0.preview.54 | Harbor link | Jul 7, 2026 | |
| 5 | DeepSeek V4 Flash | 66.4% ±2.27 SE | $49.98 | 41.4 min | No public rows | Ante0.preview.53 | Harbor link | Jul 5, 2026 | |
| 6 | MiMo V2.5 | 65.8% ±2.30 SE | $73.76 | 12.5 min | #1 same-model | Ante20260625-0824-e383a92 | Harbor link | Jun 25, 2026 | |
| 7 | MiniMax M3 | 62.1% ±2.33 SE | $121.00 | 13.9 min | #1 same-model | Ante20260623-0825-ff174ee | Harbor link | Jun 23, 2026 | |
| 8 | Qwen3.6 27B | 56.2% ±2.36 SE | Local | 61.6 min | No public rows | Ante20260701-0837-70d2aac | Harbor link | Jul 3, 2026 |
Terminal-Bench Reference
Verified Public Leaderboard
Same parameters: 89 tasks · 5 trials/task · Updated Jul 17, 2026 · Source: Terminal-Bench official verified rows
17 official rowsFor how different models perform on TB 2.1, see Vals AI's Terminal-Bench 2.1 benchmark.
| # | Agent | Model | Accuracy | Run date | |
|---|---|---|---|---|---|
| 1 | Claude CodeAnthropic | Fable 5xhigh | 83.8% ±1.16 SE | Jun 7, 2026 | |
| 2 | CodexOpenAI | GPT-5.5xhigh | 83.2% ±1.13 SE | May 1, 2026 | |
| 3 | Terminus 2Terminal-Bench | Fable 5high | 80.5% ±1.16 SE | Jun 5, 2026 | |
| 4 | Cursor CLICursor | Grok 4.5high | 79.3% ±1.46 SE | Jul 9, 2026 | |
| 5 | Claude CodeAnthropic | Opus 4.8high | 78.9% ±1.31 SE | Jul 9, 2026 | |
| 6 | CodexOpenAI | GPT-5.6 Terramax | 78.4% ±1.25 SE | Jul 11, 2026 | |
| 7 | Terminus 2Terminal-Bench | GPT-5.5xhigh | 78.0% ±1.22 SE | May 1, 2026 | |
| 8 | mini-SWE-agentPrinceton | Muse Spark 1.1xhigh | 76.2% ±1.23 SE | Jul 9, 2026 | |
| 9 | CodexOpenAI | GPT-5.6 Lunamax | 75.7% ±1.32 SE | Jul 11, 2026 | |
| 10 | Claude CodeAnthropic | Sonnet 5high | 74.6% ±1.64 SE | Jul 9, 2026 | |
| 11 | Terminus 2Terminal-Bench | Gemini 3 Prohigh | 73.9% ±1.29 SE | May 1, 2026 | |
| 12 | Claude CodeAnthropic | Opus 4.7max | 68.9% ±1.41 SE | May 1, 2026 | |
| 13 | Terminus 2Terminal-Bench | Opus 4.7max | 66.1% ±1.37 SE | May 1, 2026 | |
| 14 | Gemini CLIGoogle | Gemini 3 Prohigh | 65.8% ±1.38 SE | May 1, 2026 | |
| 15 | Gemini CLIGoogle | Gemini 3.1 Prohigh | 65.8% ±1.67 SE | May 5, 2026 | |
| 16 | Terminus 2Terminal-Bench | Gemini 3.1 Prohigh | 65.6% ±1.65 SE | May 5, 2026 | |
| 17 | Claude CodeAnthropic | GLM-5.1max | 58.6% ±1.24 SE | May 1, 2026 |
