MobiFlow - Instruction Following Success: leaderboard

MobiFlow (2026) tests mobile GUI agents on 240 real-world tasks in 20 third-party Android apps, scoring progress on task graphs fused from real user trajectories. Success rate in the instruction-following scenario, for general multimodal models in the shared framework of the paper, which marks actionable screen elements with OmniParser, in percent; higher is better.

Metric: Success rate (%) in the instruction-following scenario on MobiFlow's 240 real-world tasks in 20 third-party Android apps, scored on task-completion state graphs fused from real trajectories, at most 50 steps; general models in the paper's shared execution framework (OmniParser marks actionable elements; OpenRouter, temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-559.3#185
2Claude Opus 4.158.3#284
3Gemini 2.5 Pro41.6#280
4Gemini 3 Flash41.6#180
5Gemini 2.5 Flash33.3#466
6Grok 48.3#321

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

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