EgoIntent - Local Intent: leaderboard
Metric: Local Intent (what the person is trying to do in the current step) semantic agreement (0-100) from a DeepSeek-V4-Flash judge comparing each open-ended prediction with the human reference over the 3,014 EgoIntent micro-steps (pre-outcome visual prefix only, model-specific video or multi-frame interface; missing predictions score 0); higher is better. Source: arxiv.org. Saturation forecast: Around April 2028. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 57.99 | #123 |
| 2 | Seed 2.1 Turbo | 57.98 | #84 |
| 3 | Qwen 3 VL 32B Instruct | 45.14 | #276 |
| 4 | Gemini 3.5 Flash | 43.34 | #55 |
| 5 | Qwen 3 VL 8B Instruct | 40.86 | #401 |
| 6 | Nova 2.0 Lite | 37.98 | #408 |
| 7 | Qwen 2 VL 7B Instruct | 37.42 | #816 |
| 8 | Qwen 2.5 VL 7B Instruct | 35.41 | #643 |
| 9 | Molmo2-8B | 35.1 | #522 |
| 10 | InternVL3-8B | 34.25 | #606 |
| 11 | InternVL2-8B | 21.6 | #826 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=egointent-local-intent · How It Works · Data refreshed daily, snapshot 2026-10-11.