BioMedHop (Direct Prompting) - Path-based Reasoning: leaderboard
The Path-based Reasoning family of BioMedHop (Tan et al., 2026): recover the disease hidden inside a typed five-node SPOKE metapath from its endpoints and relation descriptions. The score is the mean of multiple-choice and alias-normalised open-answer accuracy in percent; higher is better. Direct prompting without retrieved evidence.
Metric: Accuracy (%; mean of multiple-choice accuracy and alias-normalised open-answer accuracy on Path-based Reasoning (the hidden disease on a typed five-node SPOKE metapath); stratified evaluation subsets of the 10,045 instances; direct input-output prompting with no retrieved evidence (the paper's IO baseline), temperature 0, at most 1,024 output tokens). Source: arxiv.org. Saturation forecast: Around 2030. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Turbo | 26.1 | #1002 |
| 2 | Claude 3.7 Sonnet | 25.5 | #455 |
| 3 | Gemini 2.5 Pro | 25.4 | #262 |
| 4 | Qwen 3 Next 80B A3B Instruct | 23.5 | #763 |
| 5 | Claude 3.5 Haiku | 21.4 | #1563 |
| 6 | GPT-4o Mini | 19.8 | #1627 |
| 7 | MiMo-V2.5-Pro | 19.5 | #274 |
| 8 | Qwen 3 4B | 16 | #2297 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=biomedhop-direct-prompting-path-based-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.