SRBench - ML-100K (Recall@5): leaderboard

SRBench (2026) evaluates LLMs as sequential recommenders that predict the next item a user interacts with from the full interaction history, using a structured prompt and a numeric answer extractor. This board is Recall@5 from 0 to 1 on MovieLens-100K movie ratings; higher is better.

Metric: Recall@5 (fraction) of the held-out next item among five recommendations on ML-100K, LLM sequential recommendation with the full-length interaction history and SRBench's augmented prompt and extractor; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 40.0328#410
2GPT-4.10.0293#505
3Grok 30.0269#611
4SRBench DeepSeek-v3 (checkpoint unspecified)0.0194
5GLM-3-Turbo0.0053
6Llama-2 (SRBench checkpoint unspecified)0.0037
7Llama-3.2 (SRBench checkpoint unspecified)0.0035

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=srbench-ml-100k-recall-5 · How It Works · Data refreshed daily, snapshot 2026-11-21.