OpenMobile-2 agents approach proprietary scores on phone benchmarks
The 27B model rose from 31.9 to 59.6 on real-device SPA-Bench, which the authors call the best open score.
46
AI model
Qwen3.6
- Mentions · 7d2
- Mentions · 30d3
- All mentions3
- First seenOct 13, 2026
- Last mentioned11h ago
Timeline
3 articles · newest firstCoding-agent harnesses cost more without better benchmark scores
Harness swaps changed outcomes about as often as reruns, while system prompts made some tasks cost up to three times more.
46
Interlock synthesizes completion graphs to train long-horizon agents
Fine-tuning two Qwen3.6 backbones on accepted trajectories lifts AutomationBench pass rates by 4.00-4.50 points.
40
Appears with
- Gemini 3.1 Proai model · 1
- Zhipu AIorganization · 1
- GLM-5.3-Flashai model · 1
- Deepseek-V4-Flashai model · 1
- Claude Opus 5ai model · 1
- OpenCodeproduct · 1
- OpenMobile-2ai model · 1
- mini-SWE-agentproduct · 1
- Claude Codeproduct · 1
- Nanjing Universityorganization · 1
- Qwen3.5ai model · 1
- SenseTimeorganization · 1
- Qwen3.8ai model · 1
- HY4-Previewai model · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.