JointDW - MIPLIB-NL: leaderboard
Given a mixed-integer program's description, formulation and constraint-family tables, the model names groups of constraint families that tighten the Dantzig-Wolfe bound only when convexified together; the score is hit@1, whether its first group is a true one, over 17 MIPLIB-NL models with three samples each. The MIPLIB-NL board asks a model to write such programs; this one asks it to analyse given ones.
Metric: hit@1: share of answers whose first predicted joint support is true, 17 MIPLIB-NL models, three samples each (0-1). Source: arxiv.org. Saturation forecast: Around August 2028. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 | 0.57 | #19 |
| 2 | Grok 4.20 (Reasoning) | 0.55 | #297 (Grok 4.20) |
| 3 | GPT-6.1 Sol (Medium) | 0.53 | #10 (GPT-6.1 Sol) |
| 4 | Grok 4.20 (Non-reasoning) | 0.29 | #297 (Grok 4.20) |
| 5 | DeepSeek V4 Pro (Non-reasoning) | 0.27 | #188 (DeepSeek V4 Pro) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=jointdw-miplib-nl · How It Works · Data refreshed daily, snapshot 2026-11-13.