The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestMicrosoft4B-2609
MicrosoftText / LLM

Microsoft Releases FrogNano-4B Under MIT License

The compact 4-billion-parameter text model lands on Hugging Face with a permissive license aimed at developers.

Sep 17, 2026
UpdateMIT
FrogNano-4B

Microsoft has published FrogNano-4B, a compact text-generation language model, on Hugging Face. At roughly 4 billion parameters and released under the permissive MIT license, it fits into the growing class of small models designed to run efficiently without the overhead of frontier-scale systems.

The model is a dense (non-MoE) architecture focused on text tasks. Microsoft has not published a context length or detailed benchmark specs alongside this initial drop, so practitioners will want to evaluate it against their own workloads before drawing conclusions.

Why it matters

Small models in the 1-to-7-billion-parameter range have become a workhorse category for teams that need local inference, lower latency, and predictable costs. A permissive MIT license is the key detail here: it lets developers fine-tune, embed, and ship the model commercially with minimal legal friction.

  • Roughly 4B parameters, dense architecture
  • MIT-licensed for broad commercial reuse
  • Text-only modality, available now on Hugging Face

This is the first release in the FrogNano line, so there is no prior version to compare against. Whether it carves out a niche against established small models like Microsoft's own Phi series will depend on benchmarks and community testing in the weeks ahead.

Sources

  • microsoft/FrogNano-4B-2609

    Hugging Face

    Visit
OlderTernary-Bonsai-2 packs a 27B model into 2-bit formPrism Ml · Text / LLM · 2 months agoNewerinclusionAI's Ming-Image targets graphic designinclusionAI · Text → Image · 2 months ago

Get the model

Hugging Face

Specs

Parameters4B
Context window262K tokens
Size9.3 GB
PrecisionBF16
ArchitectureQwen3_5ForConditionalGeneration
LicenseMIT
Downloads2.3K
Likes188

Can you run it?

Runs on a laptop — about 4.9 GB at 4-bit.

  • BF16 (as published)

    12 GB GPU (RTX 3060 / 4070) · 16 GB Mac

    11.4 GB
  • 8-bit

    8 GB GPU (RTX 4060) · 16 GB laptop or Mac

    7 GB
  • 4-bit

    8 GB GPU (RTX 4060) · 16 GB laptop or Mac

    4.9 GB
Your machine
GGUF builds
  • KV cache sized for 8,192 tokens of context; longer prompts need proportionally more.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Text / LLM

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More from Microsoft

All Microsoft releases →
Aion-1.0-Instruct-2.3B
Microsoft/Text / LLMRuns anywhere

Microsoft's Aion 1.0 Instruct is a compact 2.3B chat model

A small, multilingual instruction-tuned model aimed at chat and summarization lands on Hugging Face.

Nov 18, 2026
Mage-VL-Cowork
Microsoft/Vision-LanguageRuns on a laptop

Microsoft's Mage-VL-Cowork takes on computer-use tasks

A new vision-language agent model aims to operate GUIs, write code, and work the terminal.

Oct 30, 2026
VibeVoice-ASR-Streaming-7B
Microsoft/Speech → TextRuns on a laptop

Microsoft's VibeVoice ASR brings streaming speech-to-text

A 7-billion-parameter model targets real-time, multilingual transcription with an open release on Hugging Face.

Sep 2, 2026

More in Text / LLM

All Text / LLM →
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime Releases Looped 9B Multimodal Model

The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Nov 17, 2026
Hunyuan Hy4-flash
Tencent/Vision-LanguageMulti-GPU server

Tencent releases Hunyuan Hy4-flash multimodal MoE

The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.

Nov 17, 2026
Sage S2.1 Flash
Internlm/ReasoningMulti-GPU server

InternLM's Sage S2.1 Flash targets scientific reasoning

The 320B mixture-of-experts model is built to reason over scientific evidence and act inside interactive environments.

Nov 15, 2026