LifeEval (Open-Ended) - Multi-turn Interactive Collaboration: leaderboard
Metric: GPT-5 judge score against the reference answer (0 to 1 in steps of 0.25, shown times 100) on the 267 open-ended Multi-turn Interactive Collaboration questions of LifeEval, zero-shot, the model sees only the egocentric clip up to the question time (8 uniformly sampled frames for models without native video input), answer in fixed tags; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 73.92 | #91 |
| 2 | Gemini 2.5 Pro | 72.31 | #145 |
| 3 | GPT-5 Mini | 69.12 | #176 |
| 4 | Grok 4 | 67.62 | #169 |
| 5 | Gemini 2.5 Flash | 67.23 | #237 |
| 6 | GPT-4o | 59.29 | #333 |
| 7 | Qwen 3 VL 30B A3B Instruct | 52.39 | #365 |
| 8 | Qwen 3 VL 30B A3B (Thinking) | 50.41 | #338 (Qwen 3 VL 30B A3B) |
| 9 | Qwen 3 VL 8B (Thinking) | 45.78 | |
| 10 | Qwen 3 VL 8B Instruct | 42.41 | #401 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=lifeeval-open-ended-multi-turn-interactive-collaboration · How It Works · Data refreshed daily, snapshot 2026-10-11.