OmniGUI: leaderboard

OmniGUI (2026) is a step-level benchmark for smartphone GUI agents in omni-modal environments: at each of 2,579 steps in 709 expert episodes across 29 apps the agent receives a screenshot, synchronous audio and a video clip and predicts the next action, with the true action history given. This board is the step exact match in percent over all steps; higher is better.

Metric: Exact match (%): share of steps whose predicted action type and parameters (tap point inside the target element, exact typed text) both match the expert action, over all steps of OmniGUI (709 expert-demonstrated smartphone episodes, 2,579 action steps in 29 Chinese and English apps; at every step the agent receives the screenshot, a synchronous audio track and a video clip with the ground-truth action history), step-level teacher forcing, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro63.6#146
2Gemini 3 Flash61.3#180
3Gemini 2.5 Pro44.1#280
4Gemini 2.5 Flash37.8#466
5Qwen3-Omni32.3#1059
6VITA-1.512.1#3102
7MiniCPM-o-4.54.8#1251
8Baichuan-Omni-1.53.3#3189

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=omnigui · How It Works · Data refreshed daily, snapshot 2026-11-22.