CompanionBench - English - Holding Ambiguity: leaderboard
CompanionBench capability score for Holding Ambiguity: tolerating uncertainty, presence without action, space-holding and resisting the righting reflex, averaged over English-condition companionship dialogues by a cross-family LLM judge panel (CompanionBench, arXiv 2608.02046v2, Appendix Table H.1). Mean rubric score on a 1-10 scale; higher is better. Capability means correlate strongly across capabilities.
Metric: Mean judge score (1-10). Source: arxiv.org. Saturation forecast: Around February 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.7 Max | 8.75 | #85 |
| 2 | GPT-5.4 | 8.58 | #83 |
| 3 | GPT-5.5 | 8.57 | #33 |
| 4 | DeepSeek V4 Pro | 8.54 | #112 |
| 5 | Claude Opus 4.8 | 8.54 | #39 |
| 6 | Qwen 3 235B A22B 2507 Instruct | 8.52 | #315 |
| 7 | Kimi K2.6 | 8.5 | #113 |
| 8 | Kimi K2.5 | 8.48 | #152 |
| 9 | Gemma 4 31B (IT) | 8.45 | #178 |
| 10 | DeepSeek V4 Flash | 8.41 | #155 |
| 11 | Qwen 3.5 122B A10B | 8.4 | #184 |
| 12 | Claude Sonnet 4.6 | 8.38 | #94 |
| 13 | GLM-5 | 8.37 | #149 |
| 14 | GLM-5.1 | 8.37 | #109 |
| 15 | GPT-4.1 | 8.34 | #258 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=companionbench-english-holding-ambiguity · How It Works · Data refreshed daily, snapshot 2026-11-05.