AutoRO-Bench (AutoREM): leaderboard
The robust-optimization reformulation task of AutoRO-Bench (Chen et al., 2026) with AutoREM, the experience memory the paper proposes: a library of reformulation lessons distilled by DeepSeek-V4-Flash from its failed attempts and transferred unchanged to the other models. Accuracy in percent on the 64 in-distribution Random instances with a fixed DeepSeek-V4-Pro coder, mean of three runs; higher is better.
Metric: Accuracy (%; share of the 64 in-distribution Random robust-optimization instances solved correctly with the AutoREM experience memory built on DeepSeek-V4-Flash; fixed DeepSeek-V4-Pro coder, mean of three runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash (High) | 97.4 | #254 (DeepSeek V4 Flash) |
| 2 | GPT-5.4 | 96.9 | #121 |
| 3 | Qwen 3.6 Plus | 88.5 | #217 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=autoro-bench-autorem · How It Works · Data refreshed daily, snapshot 2026-11-12.