MobiFlow - Instruction Noise Completion: leaderboard
Metric: Completion rate (%) with noise added to the instruction 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 January 2027. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 71.1 |
| 2 | Claude Opus 4.1 | 68.6 |
| 3 | Gemini 3 Flash | 64.7 |
| 4 | Gemini 2.5 Pro | 56.4 |
| 5 | Gemini 2.5 Flash | 35.5 |
| 6 | Grok 4 | 8.1 |
Interactive version: theaggregate.ai/benchmark?slug=mobiflow-instruction-noise-completion · How It Works · Data refreshed daily, snapshot 2026-10-07.