Omni2Web - Instruction Recovery: leaderboard
Metric: Instruction Recovery Score (%; Track B: mean of target F1 and action F1 per instance, macro-averaged over 918 bilingual screen recordings of weakly referential webpage edit requests; the model rewrites the recording into an ordered list of explicit target and action instructions without the source HTML; one-to-one matching against reference instructions, semantic equivalence judged by Gemini 3.1 Pro; unusable outputs score zero). Source: arxiv.org. Saturation forecast: Around 2030. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 Omni Plus | 49.14 |
| 2 | Claude Opus 5 | 48.91 |
| 3 | GPT-5.6 Sol | 48.34 |
| 4 | Gemini 3.5 Flash | 47.46 |
| 5 | Seed 2.0 Lite | 46.52 |
| 6 | Gemini 3.1 Pro (Preview) | 43.19 |
| 7 | Kimi K3 | 41.95 |
| 8 | Grok 4.6 | 40.83 |
| 9 | Seed 2.1 Pro | 39.08 |
| 10 | Qwen 3.8 Max | 38.37 |
| 11 | MiMo-V2.5 | 37.31 |
| 12 | Qwen 3.8 27B | 29.84 |
| 13 | MiniMax-M3 | 29.32 |
| 14 | Muse Spark 1.1 | 25.41 |
| 15 | MiniCPM-o-4.5 | 24.9 |
Interactive version: theaggregate.ai/benchmark?slug=omni2web-instruction-recovery · How It Works · Data refreshed daily, snapshot 2026-09-26.