IG-Bench - IG-Exam: leaderboard
IdeaGene-Bench (IG-Bench, 2026) represents papers as typed Idea Genome objects aligned across 1,961 scientific lineage traces in 10 domains. IG-Exam poses 1,029 closed-form lineage-reasoning questions in 42 task types, scored by exact accuracy with every required field correct, in percent; higher is better.
Metric: Exact accuracy (%; every required field correct) on the closed-form lineage-reasoning questions of IG-Exam (1,029 instances in 42 task types over 1,961 expert-audited scientific lineage traces in 10 domains). Source: arxiv.org. Saturation forecast: Around 2029. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 23.1 | #49 |
| 2 | Qwen 3.6 Max Preview | 20.6 | #159 |
| 3 | Kimi K2 (Thinking) | 20.4 | #439 (Kimi K2) |
| 4 | Gemini 3.1 Pro (Preview) | 20.1 | #94 |
| 5 | Claude Opus 4.7 | 19.3 | #79 |
| 6 | DeepSeek V4 Pro | 17.9 | #188 |
| 7 | GLM-5.1 | 17.7 | #179 |
| 8 | MiniMax-M2.7 | 11.9 | #518 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ig-bench-ig-exam · How It Works · Data refreshed daily, snapshot 2026-11-13.