Qwen-Planner-Agent claims best result on a mobile-planning benchmark

single source· 1 articles · confidence: low · first seen 2026-09-23 20:00 UTC

What this means for you

Nothing to act on yet. There is no released model, no API, no weights and no published scores, and MobilePA-Bench is the authors' own benchmark, so the comparison cannot be checked from outside. Watch for the evaluation detail — scoring rules, device set, evaluation date — before treating the ranking as real.

A preprint posted to arXiv on 23 September 2026 describes Qwen-Planner-Agent, a system for phone planning tasks — long sequences of taps and screen reads where an early wrong step compounds. The authors say it beats every model and system they evaluated on MobilePA-Bench, their own benchmark, and improves on its base model at tool use, memory, skills and sub-agent coordination. Training combines a supervised planning cold start with online reinforcement learning in simulated and real environments; a reward-shaping method the authors call CARE is meant to cut reasoning and tool-use cost. No scores or evaluation date are given, and the paper is not peer reviewed.

Key facts

  • ·The paper is arXiv 2609.29892, posted on 23 September 2026. source
  • ·The authors report that Qwen-Planner-Agent achieves the best overall performance among all models and systems they evaluated on MobilePA-Bench. source
  • ·The authors report improvement over the base model in tool use, memory, skills and sub-agent coordination, and gains on non-mobile agentic benchmarks. source
  • ·Training combines a supervised planning cold start with online agentic reinforcement learning in hybrid (simulated and real) environments. source
  • ·The authors introduce Competence-Aware Reward-and-Advantage Engineering (CARE), described as reducing reasoning and tool-use cost while preserving task performance. source
  • ·No numeric benchmark scores, evaluation date or harness details appear in the abstract. source

What the sources say

Sources

The original reporting. Follow these — they did the work.

← the wire