FrogNano

FrogNano is Microsoft’s repository-level coding model derived from Qwen3.5-4B. Its model card dates the original release to September 22, 2026, and describes reinforcement learning on approximately 1,500 synthetic software-engineering tasks. Its intended workflow uses the five-tool Leaf harness in an isolated repository, with human review of candidate patches.

On October 17, community maintainer roman220220 uploaded a text-only mixed-precision GGUF conversion for Ollama and llama.cpp. The author’s language-model perplexity measurements do not establish coding-agent performance for this conversion. The base model card’s license section specifies Apache-2.0 while its metadata says MIT; the community card declares Apache-2.0. Users should retain applicable upstream notices and review the original terms.