FrogNano-4B-2609
FrogNano-4B-2609 is a compact coding model derived from Qwen3.5-4B and post-trained with reinforcement learning on synthetic software-engineering tasks. It generates text and structured tool calls for repository-level coding workflows, with a documented context length of approximately 131K tokens.
What it does
Developers and researchers can use it to navigate repositories, debug code, run tests, and propose patches for human review.
- Availability
- Public access to downloadable model weights, configuration files, tokenizer assets, and the model card through a Hugging Face model repository
- Pricing model
- Not established
- Deployment
- Documented rollout configuration uses SGLang on GPU servers
- Privacy
- Intended for isolated repository environments; process only code and data users are authorized to access
- Licence
- Apache License 2.0; upstream copyright and attribution notices are retained
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
6 claims- FrogNano is intended for repository-level software issue resolution from natural-language task descriptions and existing source code. [1]
• Repository-level software issue resolution from natural-language task descriptions and existing source code.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights. - The evaluated coding-agent configuration supports a combined context length of approximately 131K tokens. [1]
| Context length | Approximately 131K tokens in the evaluated FrogNano coding-agent configuration.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights. - Downloadable model weights, configuration files, tokenizer assets, and the model card are publicly accessible through a Hugging Face model repository. [1]
• Public access to downloadable model weights, configuration files, tokenizer assets, and the model card through a Hugging Face model repository.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights. - The documented rollout configuration uses SGLang on GPU servers. [1]
The upstream architecture is compatible with Hugging Face Transformers, vLLM, SGLang, and KTransformers; FrogNano’s documented rollout configuration uses SGLang on GPU servers.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights. - FrogNano is intended for human-supervised development, and generated patches require review and testing before use or deployment. [1]
Any resulting patches require human review, regression testing, and security validation before use or deployment.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights. - Apache License 2.0 applies, with applicable upstream copyright and attribution notices retained. [1]
FrogNano is derived from Qwen/Qwen3.5-4B; applicable upstream copyright and attribution notices are retained.
Microsoft. Quoted text is outside CC BY 4.0 and keeps its original rights.
Sources
1
1
MicrosoftOwner-controlled
https://huggingface.co/microsoft/FrogNano-4B-2609
Retrieved 2026-10-27 01:40 UTC
Verified
Versions
3| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 3 | 2026-10-17 19:22 UTC | evidence update | score | Current |
| 2 | 2026-10-04 07:56 UTC | material change | capability, availability | Current |
| 1 | 2026-10-03 08:48 UTC | publication | None | Current |
In briefs
2- Backburner leads 25 AI developments in week 40, 2026 Weekly brief 2026-W40
- Microsoft introduces FrogNano-4B-2609 for repository-level coding workflows Daily brief 2026-10-03
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.