Microsoft releases FrogNano-4B-2609 small agentic model

On October 3, 2026, a post on Reddit's LocalLLaMA community said that Microsoft had put FrogNano-4B-2609 on Hugging Face, describing it as an agentic model for users with limited VRAM — a model "for the GPU poor."

According to the post, FrogNano is derived from Qwen3.5-4B, a general-purpose post-trained model designed for language, reasoning, coding, agentic and multimodal tasks. FrogNano inherits Qwen3.5-4B's dense 32-layer hybrid Gated DeltaNet and gated-attention architecture, but its additional post-training is text-only and focuses on repository-level software engineering, and reinforcement learning was used in that stage.

The same post links to a Hugging Face repository, bartowski/FrogNano-4B-2609-GGUF, pointing to a GGUF-format build hosted under bartowski's account, which would make the model usable in local inference setups. Beyond the 4B figure carried in the model name, the source gives no parameter count, no context length and no benchmark results, and it does not describe how the extra post-training affects the model's behaviour compared with its Qwen3.5-4B base.

Nothing in the material indicates an accompanying Microsoft blog post, model card summary, license terms or pricing, and no release date is stated beyond the October 3, 2026 timestamp of the community post itself. A second, independent source is not present in the material, so the details above rest on that single report; the exact arrival of FrogNano-4B-2609 on Hugging Face, and any official framing from Microsoft, remain unconfirmed here.