MedMT-Bench (Text Subset): leaderboard
Metric: Accuracy (%) on the 285 text-only dialogues of MedMT-Bench's long multi-turn medical dialogues (health consultation, treatment recommendation and rehabilitation scenarios, text and image-text), where the final-turn reply is checked against fine-grained test points, judged by a Gemini-2.5-Pro evaluator at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2 | 44.91 | #236 |
| 2 | Qwen 3 32B (Thinking) | 41.75 | #424 (Qwen 3 32B) |
| 3 | GLM-4.5 (Thinking) | 40.85 | #265 (GLM-4.5) |
| 4 | Qwen 3 32B (Non-reasoning) | 34.74 | #424 (Qwen 3 32B) |
| 5 | Qwen 3 8B (Thinking) | 34.39 | #667 (Qwen 3 8B) |
| 6 | GLM-4.5 (Non-reasoning) | 34.39 | #265 (GLM-4.5) |
| 7 | Qwen 3 8B (Non-reasoning) | 33.33 | #667 (Qwen 3 8B) |
| 8 | Llama 3.1 70B Instruct | 24.91 | #548 |
| 9 | Llama 3.1 8B Instruct | 18.25 | #1018 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=medmt-bench-text-subset · How It Works · Data refreshed daily, snapshot 2026-10-11.