LogiMed-RoB - End-to-End Consistency: leaderboard
Whether an LLM's risk-of-bias assessment of a randomised controlled trial is fully consistent with the RoB 2.0 expert rule set at every level, over 860 trials. Higher is better. Source: arxiv.org/abs/2609.11185.
Metric: Fully consistent instances (%). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 45.13 | #64 |
| 2 | Gemini 3 Flash | 38.4 | #103 |
| 3 | Claude Sonnet 4.5 | 29.94 | #148 |
| 4 | GPT-5.1 | 23.8 | #145 |
| 5 | DeepSeek V3.2 | 18.85 | #210 |
| 6 | Qwen 3 32B | 1.44 | #449 |
| 7 | Llama 4 Maverick | 1.29 | #486 |
| 8 | Llama 3.3 70B | 0 | #578 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=logimed-rob-end-to-end-consistency · How It Works · Data refreshed daily, snapshot 2026-11-05.