Together AI Open Models: Model List, Licenses and Articles
About Together AI
Together AI is a company offering a cloud for running open models. The card of its togethercomputer organization on Hugging Face says it is building a decentralized cloud for working efficiently with large foundation models.
Rather than releasing large foundation models of its own, it publishes experimental models built on other companies’ models and helper models that speed up inference.
What It Releases
- Tev1 (September 2026): experimental 0.8B and 4B “decision models". According to the model card, they are fine-tunes of Qwen3.5-4B (Qwen3.5-0.8B for the 0.8B) trained to choose one option from a structured state, question and list of choices. Its official blog shows how to train such a classifier for about US$17 (we have an article on it).
- Aurora-Spec (February 2026): helper models (EAGLE3 draft models) that speed up inference by speculative decoding, for MiniMax-M2.1, M2.5 and Qwen3-Coder-Next. According to the model card, they were trained from random initialization with Aurora, an inference-time training framework.
- Earlier research models include Dragonfly (reads images) and MoAA.
Licenses
Aurora-Spec is Apache-2.0, M1-3B is MIT, and Llama-based models use the original Llama license. Some models, such as Tev1, have no license field. Read each model’s license text before use. This summary is not legal advice.
Running Them Locally
- Tev1 is small (0.8B and 4B) and easy to try locally, but it is focused on choosing among options, not general chat.
- Aurora-Spec is not used on its own. It is a helper model paired with the matching main model for speculative decoding in vLLM, SGLang and similar engines.
- There are no official GGUF builds (as of 2026-09-26).
Sources: the togethercomputer organization card; model cards of togethercomputer/Tev1-4B-experimental and togethercomputer/Aurora-Spec-Minimax-M2.5; the model list of the togethercomputer organization on Hugging Face (all as of 2026-09-26).
Our Coverage and Data
Local Model Watch has published 2 article(s) on Together AI’s own models, 0 on third-party fine-tunes and quantized builds based on them, and 4 from its official blog. 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
| Item | Value |
|---|---|
| Hugging Face organization | togethercomputer |
| Models on Hugging Face | 38 |
| Articles on its own models | 2 |
| Articles on derived models | 0 |
| Articles from its official blog | 4 |
Recent Models on Hugging Face
The publisher lists 38 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.
→ Scroll horizontally to see all columns
| Released | Model | Task | Parameters | License | Official variants |
|---|---|---|---|---|---|
| 2026-09-26 | togethercomputer/Tev1-0.8B-experimental | text generation | 873M | — | — |
| 2026-09-24 | togethercomputer/Tev1-4B-experimental | text generation | 4.7B | — | — |
| 2026-02-20 | togethercomputer/Aurora-Spec-Minimax-M2.5 | text generation | 858M | apache-2.0 |
— |
| 2026-02-05 | togethercomputer/Aurora-Spec-Minimax-M2.1 | text generation | 858M | apache-2.0 |
— |
| 2026-02-03 | togethercomputer/Aurora-Spec-Qwen3-Coder-Next-FP8 | text generation | 519M | apache-2.0 |
— |
| 2025-05-02 | togethercomputer/M1-3B | text generation | 3.4B | mit |
— |
| 2024-12-06 | togethercomputer/gemma-2-9b-it-MoAA-DPO | text generation | 9.2B | — | — |
| 2024-12-06 | togethercomputer/Llama-3.1-8B-Instruct-MoAA-DPO | text generation | 8.0B | — | — |
| 2024-10-11 | togethercomputer/Llama-3.1-8B-Dragonfly-Med-v2 | vision-language | — | llama3.1 |
— |
| 2024-10-11 | togethercomputer/Llama-3.1-8B-Dragonfly-v2 | vision-language | — | llama3.1 |
— |
| 2024-09-23 | togethercomputer/gemma-2-9b-it-MoAA-SFT | text generation | 9.2B | — | — |
| 2024-09-20 | togethercomputer/Llama-3.1-8B-Instruct-MoAA-SFT | text generation | 8.0B | — | — |
| 2024-06-03 | togethercomputer/Llama-3-8B-Dragonfly-Med-v1 | vision-language | — | llama3 |
— |
| 2024-06-03 | togethercomputer/Llama-3-8B-Dragonfly-v1 | vision-language | — | llama3 |
— |
| 2024-02-25 | togethercomputer/evo-1-8k-base | text generation | 6.5B | apache-2.0 |
— |
From the Hugging Face API. Last changed 2026-09-27 (JST). Collected by code.
By Model Family
→ Scroll horizontally to see all columns
| Family | Own-model articles | Derived-model articles | Smallest VRAM tier | Latest |
|---|---|---|---|---|
| Qwen3.5 | 2 | 0 | — | 2026-09-26 |
Articles on Its Own Models
| Published | Model | Type | Article |
|---|---|---|---|
| 2026-09-26 | togethercomputer/Tev1-0.8B-experimental | New Models | Tev1-0.8B-experimental Text Generation Model: 4GB+ VRAM |
| 2026-09-24 | togethercomputer/Tev1-4B-experimental | New Models | Tev1-4B-experimental Text Generation Model: 4GB+ VRAM, GGUF Builds |
From Its Official Blog
- 2026-09-24 Train Your Own Jev-style Classifier for $17 with Together AI
- 2026-09-23 Together AI Announces Canary Rollouts for Zero-Downtime Updates
- 2026-09-19 Scaling Coding Agent Traffic with GLM-5.2 and Dedicated Inference
- 2026-09-10 Together AI Explains Open-Source AI Stack and MIGHT Stack
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.