RecRM-Bench - Query-Item Relevance: leaderboard

RecRM-Bench (Meituan, 2026) tests models as reward models for an agentic recommender system, on real query-response logs. This board covers the three-level relevance score between the query and a recommended item; the score is the percent of judgements that match the gold label; higher is better.

Metric: Accuracy (%) of the three-level relevance score (irrelevant, weakly relevant, fully relevant) the model assigns to a query-item pair, 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 February 2027. 4 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 MaxNon-reasoning76.64#415
2LongCat-Flash-Thinking75.97#749
3DeepSeek V3.2Thinking75.22#419
4LongCat-Flash-Chat73.18#818

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-query-item-relevance · How It Works · Data refreshed daily, snapshot 2026-11-21.