Hugging Face (harness-hub team) · Coding Agent LoRA
Qwen3.8-27B-pi-adapted
Hugging Face's harness-hub team has released Qwen3.8-27B-pi-adapted, a rank-32 LoRA adapter for Qwen3.8-27B trained with reinforcement learning (AsyncGRPO from TRL) to work inside the pi coding agent. Self-reported results on pi with thinking off show SWE-bench Pro (hard 51) rising from 39.2% to 64.7% and Terminal-Bench 2.1 from 39.5% to 55.8%, with median turns on SWE-bench Pro falling from 59 to 23. The adapter is Apache-2.0 licensed, works only on Qwen3.8-27B, and can be served with vLLM or llama.cpp; the evaluation sets are small and results are reported by the authors.
Open the source ↗Similar Agents & Automation releases
- Claude Code Code Mode
Anthropic (reported by MarkTechPost) · Agent Code Mode · Week 48, 2026
- TokenEconomizer
Microsoft Research (with UC Berkeley and Microsoft Security) · Coding Agent Plugin · Week 48, 2026
- ESCAPEMENT
Leonxlnx · Agent-made Blender film · Week 47, 2026
- Cindy
makecindy · Multi-device agent client · Week 47, 2026
- OpenMontage
OpenMontage (Calesthio AI Labs) · Agentic video studio · Week 47, 2026
- Impeccable
Paul Bakaus (pbakaus) · Design skills for agents · Week 47, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.