Issue 2026-11-17 Β· Voices

clem πŸ€— @ClementDelangue

Introducing Qwen3.8-27B-pi-adapted πŸ€— We adapted Qwen's model specifically for the pi coding agent using reinforcement learning and a LoRA adapter. The results, measured as pass@1: πŸš€ SWE-bench Pro (Hard 51): 29.4% β†’ 62.7% πŸ’» Terminal-Bench 2.1: 39.5% β†’ 55.8% ⚑ 55–62% fewer tool calls per task This is the power of open models: you can adapt them to YOUR tools and YOUR use cases. Let’s build! πŸ’› Open weights: https://t.co/aH4cl2qnED Methodology: https://t.co/avSfF93K4V

X17 h ago
Read original β†—