FrogNano details: adaptive on-policy synthetic RL envs train a 4B coding model to 61.5%
sivareddyg · x · 2026-09-09
The FrogNano team shares details: Qwen3.5-4B post-trained purely with RL on synthetic TaskPilot tasks — zero distillation, zero teacher SFT, just 5 iterations × 300 tasks (stopped at 1500) — reaching 61.5% on SWE-bench Verified. The key is 'on-policy data': generating synthetic RL environments targeted at the current policy's learnability zone and adapting them continuously from rollout feedback, with surprising generalization to TB2, SwePro and PatchEval.
More from coding & agent
- Hugging Face ships hf CLI updates: deduped cache saves 19GB, faster buckets, retryable GPU jobs — yb2698 · 2026-10-26
- Fern 0.1 debuts: a model built to audit training data and debug agent traces for AI builders — bneyshabur · 2026-10-26
- One prompt, full Blender animation: dev says he never touched Blender once — iamfakhrealam · 2026-10-26
- Claude Code finds root backdoor in Honda Civic head unit, then debloats it — IndraVahan · 2026-10-26
- Claude Code + Monid MCP builds a $12K/year B2B prospecting engine on live data — aryanXmahajan · 2026-10-26
- Karpathy shares CLAUDE.md rulebook to stop AI coding agents from piling up tech debt — mdancho84 · 2026-10-26