Agent2 RL-Bench - DeepSearchQA: leaderboard

Agentic post-training engineering on Agent^2 RL-Bench (2026): a coding agent driven by the model has 12 hours in an isolated GPU workspace to post-train Qwen3-8B-Base, and the grading server keeps the best checkpoint score of the run. DeepSearchQA accuracy of the trained model with a search tool, in percent; higher is better.

Metric: DeepSearchQA accuracy (%) on 200 held-out questions with a search tool and an LLM judge of the Qwen3-8B-Base model after the coding agent (driver LLM in a CLI scaffold) post-trains it for 12 hours in an isolated GPU workspace, best score the grading server returned over the run, single run; higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.623#113
2GPT-5.421#135
3Claude Sonnet 4.518.5#261
4GPT-4o17#751
5Gemini 2.5 Flash13#463
6GPT-5.212#185

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=agent2-rl-bench-deepsearchqa · How It Works · Data refreshed daily, snapshot 2026-11-21.