MedMeta: leaderboard
MedMeta (Ha, Favre and Portet, 2026) asks a model to write the main conclusion of a medical meta-analysis given its title and the abstracts of all its primary studies (81 meta-analyses published 2018-2025, 24 specialties). A panel of three LLM judges (Gemini 2.5 Pro, o4-mini, Qwen3-235B) scores each conclusion 0-5 for semantic equivalence to the original; higher is better.
Metric: LLM-judge score (0-5; mean of Gemini 2.5 Pro, o4-mini and Qwen3-235B judges rating semantic equivalence of the generated conclusion to the meta-analysis's own, 81 meta-analyses; Golden-RAG: the title and all primary-study abstracts). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 3.16 | #452 |
| 2 | O4 Mini | 2.79 | #354 |
| 3 | Qwen 3 8B (Non-reasoning) | 2.72 | #1838 (Qwen 3 8B) |
| 4 | Gemma 3 27B | 2.58 | #1638 |
| 5 | Qwen 3 8B (Thinking) | 2.56 | #1838 (Qwen 3 8B) |
| 6 | DeepSeek R1 0528 Qwen3 8B | 2.55 | #2101 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medmeta · How It Works · Data refreshed daily, snapshot 2026-11-12.