SciR - Induction Hard (Obfuscated): leaderboard
Accuracy on the 200 hard rule-induction tasks over DrugBank drug-drug interaction facts, rendered as obfuscated multi-document scientific discourse (lab notes, drug labels, records and tables), from SciR (2026), a scientific-reasoning benchmark generated from formal objects. The model reasons by chain of thought at temperature 0 and answers are checked against the generating object. Higher is better.
Metric: Accuracy (%) on the 200 hard-tier SciR induction tasks (rule induction over DrugBank drug-drug interaction facts), rendered as obfuscated multi-document scientific discourse (Obf: lab notes, drug labels, records and tables); the model reasons with chain-of-thought directly from the task text, temperature 0; answers are checked against the generating formal object; higher is better. Source: arxiv.org. Saturation forecast: Around June 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 | 26 | #473 |
| 2 | GPT-4o | 25.5 | #741 |
| 3 | OLMo 3.1 32B Instruct | 23.5 | #2384 |
| 4 | Llama 3.3 70B Instruct | 18.5 | #1357 |
| 5 | Qwen 3 30B A3B 2507 Instruct | 16 | #1089 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=scir-induction-hard-obfuscated · How It Works · Data refreshed daily, snapshot 2026-11-13.