Latenode

Shanghai AI Lab releases Atria Dawn weights with 512K context

Builders get an open-weight long-context agent model they can test through familiar API formats, with a local deployment path if they have the hardware. The main constraints are deployment cost and the lack of multimodal input.

Shanghai AI Lab releases Atria Dawn weights with 512K context

Shanghai AI Lab has published Atria Dawn as an open-weight release, with hosted access and local deployment both supported. The Atria Dawn access page shows OpenAI-compatible and Anthropic-compatible API options for the model.

On the Hugging Face model card, the full release is listed with a 512K context window, while preview variants are capped at 256K. The same card says Atria Dawn is built on a 744B-parameter MoE GLM-5.2 foundation model, and that the code and model weights are available under the MIT License.

For builders, that means a long-context, permissively licensed agent model that can sit behind familiar API patterns instead of a single vendor stack. The practical limit is deployment cost: a model at this size is not a lightweight self-hosting option, and the release is text-only, with image inputs explicitly rejected.

When models and APIs shift keep the workflow running

Connect the models and apps from stories like this in one scenario. If a vendor changes routing, pricing, or availability, you update the workflow — you don't start over.

Start Free

Free forever plan. No credit card required.