Swallow (Institute of Science Tokyo) Open Models: Model List and Licenses

September 27, 2026

About Swallow (Institute of Science Tokyo)

According to its organization card, Swallow is a family of large language models developed mainly by members of the Okazaki and Yokota Laboratories at the Institute of Science Tokyo (formerly Tokyo Institute of Technology). It is released on the tokyotech-llm organization on Hugging Face.

Swallow is known for continually pre-training strong overseas models in Japanese to make them strong in both Japanese and English, with training data designed to keep math and coding ability. Its models are also the base for other Japanese publishers’ work, such as CyberAgent’s CAT series.

What It Releases

• Qwen3-Swallow v0.2 (February 2026): 8B, 30B-A3B (MoE) and 32B. According to the model card, they are Japanese-English models built from Qwen3 through continual pre-training (CPT), SFT and reinforcement learning with verifiable rewards (RLVR), with reasoning on par with the original Qwen3 and better on some tasks. Weights from each stage (CPT, SFT, RL) are released too.
• GPT-OSS-Swallow v0.1 (February 2026): 20B and 120B based on OpenAI’s gpt-oss, made the same way as Qwen3-Swallow.
• Medical models (June 2026): Medical-*-Swallow, based on Qwen3-Swallow and GPT-OSS-Swallow.
• Earlier lines such as Llama-3.1-Swallow and Swallow-MX (Mixtral-based) are also available.

Licenses

Qwen3-Swallow, GPT-OSS-Swallow and the medical models are Apache-2.0. Llama-3.1-Swallow lists both the Llama 3.3 and Gemma licenses (inheriting the terms of the models used in training). Read each model’s license text before use. This summary is not legal advice.

Running Them Locally

• There are no official GGUF builds (as of 2026-09-26). Instead, Qwen3-Swallow has official AWQ (INT4) builds, and GPT-OSS-Swallow has MXFP4 builds like the original gpt-oss.
• On a consumer GPU, the Qwen3-Swallow 8B or 30B-A3B are the candidates. The 30B-A3B is an MoE with about 3B active.

Sources: the tokyotech-llm organization card; model cards of tokyotech-llm/Qwen3-Swallow-8B-RL-v0.2 and tokyotech-llm/GPT-OSS-Swallow-20B-RL-v0.1; the model list of the tokyotech-llm organization on Hugging Face (all as of 2026-09-26).

Our Coverage and Data

Local Model Watch has not published an article on Swallow (Institute of Science Tokyo)’s models yet. The model list below is collected by code from Hugging Face every day. Part of our publisher index; model families link to their family pages.

At a Glance

Recent Models on Hugging Face

The publisher lists 138 models in total on Hugging Face. Below are up to 15 of the most recent generative models (text, image, video, audio and so on), the scope of this site, per organization. Classifiers, feature extractors, research components and repositories without a model card are omitted, and the publisher’s own quantized builds and helper variants are folded into the row of the original model. Parameter counts are left blank for repositories that pack weights into integer types, where Hugging Face’s count is not the real parameter count.

From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.

Last updated 2026-09-27 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log and the Hugging Face API.