CAMP-II Frozen Coding Baselines - Terminal-Bench 2.1: leaderboard
CAMP-II Frozen Coding Baselines - Terminal-Bench 2.1: Task success (%). CAMP-II arXiv 2611.14009v1: unmodified Qwen3.5-4B/9B and frozen Qwen3.8-27B teacher on CAMG-II five-domain held-out panels. Same-generation coding baselines also retained on Terminal-Bench 2.1 and SWE-Bench Pro V2. Trained memory policies (including CAMP-II) lack released checkpoint identities: code is promised at camera-ready, not currently released. Frozen baseline scores retain the native boundary-write continuation harness; effort and decoding unreported.
Metric: Task success (%). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 35B A3B | 33.8 | #480 |
| 2 | Qwen 3.5 9B | 23.1 | #803 |
| 3 | Qwen 3.5 4B | 23.1 | #1180 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=camp-ii-frozen-coding-baselines-terminal-bench-2-1 · How It Works · Data refreshed daily, snapshot 2026-11-12.