iAgentBench (Reflexion): leaderboard
Metric: Reflexion accuracy (%) on a random 500-question sample of iAgentBench, user-like open-domain questions seeded from high-traffic GDELT topics whose answers combine evidence across several themes of the retrieved web corpus, graded with the SimpleQA grading procedure; the model retrieves the first page of SearXNG results and revises its answer with agentic self-reflection; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 68.2 | #138 |
| 2 | Llama 4 Maverick | 62.8 | #451 |
| 3 | Gemma 3 27B | 57 | #596 |
| 4 | Mistral Large 3 | 56.4 | #388 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=iagentbench-reflexion · How It Works · Data refreshed daily, snapshot 2026-10-11.