Agent2 RL-Bench - ALFWorld: 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. ALFWorld success rate of the trained model on evaluation episodes, in percent; higher is better.

Metric: ALFWorld success rate (%) on 134 evaluation episodes 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: Estimated already saturated. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.697.76#113
2GPT-5.490.93#135
3GPT-5.287.31#185
4Claude Sonnet 4.515.67#261
5Gemini 2.5 Flash11.19#463
6GPT-4o9.7#751

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-alfworld · How It Works · Data refreshed daily, snapshot 2026-11-21.