FrogNano-4B-2609

Microsoft FrogNano-4B: compact repository-level coding agent (Qwen3.5-4B, RL-only post-training). Thinks, MIT.

🚀 Overview

GGUF build of microsoft/FrogNano-4B-2609, Microsoft’s 4B repository-level coding agent derived from Qwen3.5-4B and post-trained only with reinforcement learning on synthetic software-engineering tasks. It ships with its vision projector, so image input works in Ollama.

Microsoft evaluated it with its five-tool Leaf harness; the vision tower is inherited from Qwen3.5-4B and was not post-trained.

🎯 Key Features

• Repository-level coding agent in 4B
• Post-trained purely with RL — no stronger-model trajectories
• Thinking model, 262K-token context
• MIT license

🏷️ Available Versions

💻 Quick Start

ollama run richardyoung/frognano-4b-2609
# then drag an image file into the prompt, or send it in the API "images" field

🛠️ Use Cases

• Local coding agents
• Code review and repo Q&A
• Research on RL-trained agents

✅ Verified

Tested through Ollama before upload (Q4_K_M and Q8_0) on a synthetic invoice with the prompt None; Q4_K_M returned: ‘INVOICE 4471 / Date: 2026-09-28 / Widgets x3 ……. \(42.00 / TOTAL: \)86.20’.

🔧 Technical Details

• Base Model: microsoft/FrogNano-4B-2609
• Parameters: 4.3B
• Context Length: 256K tokens
• Quantization: GGUF via llama.cpp with an importance matrix (calibration_datav3); vision projector at F16
• License: MIT (inherited from the base model)

🙏 Acknowledgments

• Base Model: Microsoft
• Quantization: llama.cpp

Built & maintained by Richard Young · DeepNeuro