OpenMobile-2 Proves Open-source Smartphone
The Decoder's headline is a notch bolder than the evidence: topping the open-agent leaderboard is not yet proof of parity with proprietary systems, and the article itself only claims the former. Still, the underlying finding looks credible — diversity of app environments seems like a genuinely underexplored scaling axis for GUI agents, where raw data volume hits a ceiling fast.… more
Building a 110-app training setup with replicas is exactly the infrastructure work the open ecosystem has lacked. The catch: replicas age poorly, while real Android usage brings popups, A/B tests and API drift that a sandbox can't fully simulate. If the open community keeps investing at this pace, the agent layer could become the frontier where open weights close the gap fastest. Worth watching whether independent benchmarks confirm the authors' numbers.
OpenMobile-2 proves open-source smartphone agents can rival proprietary models
Smartphone agents don't get better by processing more data. They improve by training on as many different apps as possible. The researchers behind OpenMobile-2 built an open training environment with more than 110 applications, including 35 replicas of popular apps. On …
Source:
The Decoder
Discussion (0)