ShatterMed-QA (Chinese, Hard): leaderboard
Metric: Accuracy (%) on the 327 Chinese Hard multiple-choice clinical vignettes synthesized from a hub-pruned (k-shattered) medical knowledge graph, with the bridging entity masked and a hard distractor derived from its sibling node; zero-shot, closed-book; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 96.94 | #346 |
| 2 | Grok 4.1 Fast (Reasoning) | 96.64 | #208 (Grok 4.1 Fast) |
| 3 | Grok 4 Fast (Reasoning) | 96.64 | #242 (Grok 4 Fast) |
| 4 | GPT-5 Mini | 96.33 | #176 |
| 5 | Grok 4.1 Fast (Non-reasoning) | 96.33 | #208 (Grok 4.1 Fast) |
| 6 | Grok 4 Fast (Non-reasoning) | 96.02 | #242 (Grok 4 Fast) |
| 7 | Qwen3-14B-Base | 92.35 | #610 |
| 8 | InternLM3-8B-Instruct | 89.3 | #847 |
| 9 | GPT-5 Nano | 86.54 | #415 |
| 10 | GPT-4.1 Nano | 84.71 | #716 |
| 11 | Yi-1.5-9B | 80.73 | #1230 |
| 12 | Gemma 2 9B | 79.51 | #940 |
| 13 | c4ai-command-r7B-12-2024 | 74.31 | #1145 |
| 14 | Llama 3.1 8B | 67.89 | #1139 |
| 15 | Falcon3-10B-Base | 63.61 | #1191 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=shattermed-qa-chinese-hard · How It Works · Data refreshed daily, snapshot 2026-10-11.