​
New Features

  • Template Library Added to Console

    1. We’ve added a Template Library to the console, featuring templates created by both the official team and the user community.
    2. In addition to deploying and saving templates, users are encouraged to share templates they like by copying the link with a single click.
    3. To contribute your own templates, simply set the visibility to Public when creating one. Public templates will appear in the Template Library, and those that gain popularity may be featured.

    Updated time: April 7, 2025

  • New Model: meta-llama/llama-4-maverick-17b-128e-instruct-fp8

    Novita has deployed Llama 4 Maverick 17B Instruct (128E) , a high-performance multimodal language model developed by Meta using a Mixture-of-Experts (MoE) architecture. The model dynamically activates 17 billion parameters per inference from a total of 400 billion parameters, distributed across 128 expert modules. It supports text and image inputs in 12 languages, and produces text and code outputs. Designed specifically for vision-language tasks, the model has been instruction-tuned to handle assistant-style interactions, image reasoning, and a wide range of general multimodal applications.

    Updated time: April 7, 2025

  • New Model: meta-llama/llama-4-scout-17b-16e-instruct

    Novita has deployed Llama 4 Scout 17B Instruct (16E), a Mixture-of-Experts (MoE) language model developed by Meta. The model dynamically activates 17 billion parameters per inference from a total of 109 billion parameters, utilizing 16 expert modules each time. It supports multimodal inputs (text and images) and generates text or code outputs in 12 languages. Designed for intelligent assistant interactions and visual reasoning tasks, It also supports a 10 million-token context window, and was trained on a massive 40-trillion-token dataset.

    Updated time: April 7, 2025

​
Improvements

  • Improved Instance Status Visibility

    Two new intermediate statuses—“Starting” and “Stopping”—have been introduced to provide more granular visibility during instance state transitions.

    Updated time: April 10, 2025

Was this page helpful?