Computer Anthology Terminal Tasks (Terminus-2): leaderboard
Vetto's Computer Anthology Terminal Tasks v1.0 — 100 held-out terminal jobs across 17 categories and 11 runtimes, weighted toward data science, ML, optimization and model training, each graded by its own sandboxed verifier (no LLM judge). This slice runs every model under the Terminus-2 scaffold; pass@1 over 5 trials per task.
Metric: pass@1 (%). Source: vetto.ai. Saturation forecast: Around February 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Medium) | 67 | #6 (GPT-6 Astra) |
| 2 | Claude Fable 5.1 (High) | 62.4 | #10 (Claude Fable 5.1) |
| 3 | Claude Opus 5 (High) | 62 | #16 (Claude Opus 5) |
| 4 | Gemini 3.8 Flash (Medium) | 51.8 | #30 (Gemini 3.8 Flash) |
| 5 | GPT-5.6 Sol (Medium) | 51.6 | #22 (GPT-5.6 Sol) |
| 6 | Claude Fable 5 (High) | 50.8 | #20 (Claude Fable 5) |
| 7 | Grok 4.6 (High) | 49.2 | #29 (Grok 4.6) |
| 8 | Gemini 3.7 Flash (Medium) | 46 | #24 (Gemini 3.7 Flash) |
| 9 | Kimi K3 (Max) | 40 | #35 (Kimi K3) |
| 10 | Muse Spark 1.1 | 39 | #55 |
| 11 | GPT-5.5 (Medium) | 38.6 | #33 (GPT-5.5) |
| 12 | Hy4 preview (High) | 33 | #45 (Hy4 preview) |
| 13 | DeepSeek V4 Pro (0813) (High) | 32.6 | #58 (DeepSeek V4 Pro (0813)) |
| 14 | GLM-5.2 (High) | 31.4 | #80 (GLM-5.2) |
| 15 | Claude Opus 4.8 (High) | 29.8 | #39 (Claude Opus 4.8) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=computer-anthology-terminal-tasks-terminus-2 · How It Works · Data refreshed daily, snapshot 2026-11-05.