TreeFlex Reference - BrowseComp-Plus - Flat ReAct 32k - Mean@8: leaderboard
Table 2; 150-question FoldAgent test split, eight samples per question; fixed Flat ReAct 32k active context, no branching; source label base means before task RL, explicitly Instruct-2507 checkpoints; no RL or alternate harness scores pooled. Metric: Accuracy (%). Higher is better. Source: https://arxiv.org/html/2610.20334v1
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around January 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 30B A3B 2507 Instruct | 42.6 | #481 |
| 2 | Qwen 3 4B 2507 Instruct | 36.2 | #743 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=treeflex-reference-browsecomp-plus-flat-react-32k-mean-8 · How It Works · Data refreshed daily, snapshot 2026-10-16.