Open-source Buddy benchmark: Sonnet 5.5 fastest at driving a real Android phone at 25s
3amae404 · reddit · 2026-10-18
A developer built Buddy, an open-source Android app that lets Claude Code, Codex or OpenCode operate a real phone, and ran a small benchmark on a OnePlus 12: 5 everyday tasks, 2 runs per model, apps reset between runs, no approvals.
Median time and accuracy:
- Sonnet 5.5: 25s, 10/10
- Opus 5.5: 27s, 10/10, fewest steps
- GPT-6 Luna: 29s, 7/10 (twice reported the day's high instead of current temp)
- Haiku 5.5: 30s, 10/10
- GPT-6.1 Sol: 36s, 10/10
- Space Bunny (free): 43s, 10/10, once mistyped a number and self-corrected
- Big Pickle (free): 112s, 10/10
Notable behaviors: Opus never tapped through Maps — it opened a directions deep link and read the answer in 15s; some models skipped Google and fetched weather from wttr.in. Small sample, raw data shared in comments.
Related event: Buddy Tests AI Agents Controlling Real Android Phones(2 posts)→
More from coding & agent
- Building a multi-agent shared knowledge base with two GitHub repos — and one memory pitfall — gregbarbosa · 2026-10-18
- AEGIS Trust: Open-Source VS Code Extension to Rein In Drifting AI Coders — Michael0647 · 2026-10-18
- Fully agentic MV pipeline: Claude + MiniMax H3 renders 28 scenes with 0ms sync drift — Cheap_Credit_3957 · 2026-10-18
- Dev builds seamless laptop/web/mobile dev env with Claude Code self-hosted runners + Tailscale — davidcrawshaw · 2026-10-18
- SOC Prime launches LogTotal: rule-first log analysis with AI interpretation at the end — cyb3rops · 2026-10-18
- Thorsten Ball: A 'wrong' code push matters far less now that AI makes changes cheap — sloppenheimer · 2026-10-18