←Story detail

Trending story

Research team releases open-source mobile phone agent model OpenMobile-2

1 reports1 story updatesUpdated Oct 25 14:55

Understand this story

AI overview

Researchers from Nanjing University, SenseTime and other institutions have released an open-source mobile phone agent model, OpenMobile-2, comprising two agent models: a 9-billion-parameter model based on Qwen3.5 and a 27-billion-parameter model based on Qwen3.6, along with the complete training environment, data and reproduction instructions. The team rebuilt 35 applications including Taobao, JD.com, Didi, Meituan, Slack and Google Drive, with the practice environment covering more than 110 applications, generating about 11,600 example workflows and more than 2,000 practice tasks. The larger model rose from 67.1 to 79.9 on AndroidWorld and from 31.9 to 59.6 on SPA-Bench on real Android devices, which the researchers say is the best result among open-source models. There are currently no further developments.

AI-generated from public reports

Latest development · Report

OpenMobile-2 open-source mobile agent matches closed-source models on real-device benchmarks

Developments

1 confirmed developments

  1. Research team releases open-source mobile phone agent model OpenMobile-2

    Source:The Decoder

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Oct 25

  1. The Decoder

    OpenMobile-2 open-source mobile agent matches closed-source models on real-device benchmarks

    Researchers from Nanjing University, SenseTime, and other institutions have released OpenMobile-2, consisting of two agent models—a 9B model based on Qwen3.5 and a 27B model based on Qwen3.6—and have publicly released the training environment, data, and reproduction instructions. The team recreated 35 applications including Taobao, JD.com, DiDi, Meituan, Slack, and Google Drive; the practice environment covers more than 110 applications and generated about 11,600 example workflows and more than 2,000 practice tasks. The larger model rose from 67.1 to 79.9 points on AndroidWorld and from 31.9 to 59.6 points on SPA-Bench on real Android devices; the researchers say this is the best result among open-source models.

Related stories

  1. Related storiesDeepSeek open-sources inference framework DeepInfer
  2. Related storiesCohere discloses that the Qwen3.8 model attempted to exfiltrate files in a sandbox
  3. Related storiesOpenAI says GPT-Rosalind offers new leads for neurodegenerative disease drug discovery
  4. Related storiesNvidia releases Nemotron 3.5 Super VL and Bolt models