GroupAssistBench - L1 M3 F1: leaderboard
GroupAssistBench - L1 M3 F1: Source-native recall/F1 or mean rubric score (0-1). GroupAssistBench arXiv 2611.04867v1: 1374 recall and 1021 memory-use instances from 50 group histories. L1 Atom Recall/F1 and L2 rubric scores (0-1) remain separate, M1/M2/M3 and pooled scores distinct. Four full-context models and GPT-5.6-terra with ten retrieval/memory methods keep native context access and memory shape. GPT Terra compatibility temperature=1; other generation limits/effort follow adapters and are unreported here. Rank/context-token columns not capability; no judge score substituted for actor identity.
Metric: Source-native recall/F1 or mean rubric score (0-1). Source: arxiv.org. Saturation forecast: Around 2029. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra | 0.36 | #60 |
| 2 | DeepSeek V4 Flash | 0.35 | #254 |
| 3 | Claude Sonnet 5 | 0.35 | #108 |
| 4 | Qwen 3.5 9B | 0.28 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=groupassistbench-l1-m3-f1 · How It Works · Data refreshed daily, snapshot 2026-11-12.