MedConclusion: leaderboard
Metric: Semantic similarity (judge score 0-100): the model writes a formal academic conclusion (prompt A, no length or style constraint) from the non-conclusion sections of a PubMed structured abstract; a GPT-5.4-mini judge scores it 0-100 against the author-written conclusion, averaged over a random 30K-abstract subset of MedConclusion; temperature 0, at most 1,024 new tokens; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 73.22 |
| 2 | Gemini 3.1 Pro (Preview) | 71.87 |
| 3 | MiniMax-M2.1 | 71.21 |
| 4 | Gemma 3 27B (IT) | 71.03 |
| 5 | GLM-4.6V | 70.86 |
| 6 | Llama 3.1 8B Instruct | 70.53 |
| 7 | Qwen 3 4B 2507 Instruct | 69.8 |
| 8 | Kimi K2 (Thinking) | 69.79 |
| 9 | DeepSeek V3.2 (Non-reasoning) | 69.47 |
| 10 | Gemma 2 9B (IT) | 69.31 |
| 11 | Qwen 2.5 VL 7B Instruct | 68.96 |
| 12 | DeepSeek R1 | 68.93 |
| 13 | Qwen 2.5 7B Instruct | 66.87 |
| 14 | Llama 3.2 1B Instruct | 54.17 |
Interactive version: theaggregate.ai/benchmark?slug=medconclusion · How It Works · Data refreshed daily, snapshot 2026-10-07.