RecRM-Bench - Instruction Following: leaderboard
RecRM-Bench (Meituan, 2026) tests models as reward models for an agentic recommender system, on real query-response logs. This board covers the instruction-following compliance score under expert rubrics; the score is the percent of judgements that match the gold label; higher is better.
Metric: Accuracy (%) of the overall instruction-following compliance score the model assigns to a recommender response under the expert rubrics (role, format, process, constraint, content quality, style), zero-shot, the model acting as a reward model for an agentic recommender on RecRM-Bench (real Meituan query-response logs), its parsed score compared with the gold label; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 5 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | LongCat-Flash-Chat | 64.84 | #818 | |
| 2 | GPT-4.1 | 55.47 | #505 | |
| 3 | LongCat-Flash-Thinking | 43.75 | #749 | |
| 4 | Qwen 3 Max | Non-reasoning | 36.8 | #415 |
| 5 | DeepSeek V3.2 | Thinking | 35.29 | #419 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=recrm-bench-instruction-following · How It Works · Data refreshed daily, snapshot 2026-11-21.