Meta has announced Muse Code (beta), MSL's first coding agent, alongside Muse Spark 1.2, an upgraded AI model designed to power the coding tool and support more advanced software development tasks.
Mark announced Muse Code and Muse Spark 1.2, following the release of Muse Spark 1.1 last month. The latest launch highlights the rapid pace of development across Meta's AI technology stack, with both products now available through expanded global access via the Meta Model API.
Muse Code is a terminal-based coding tool designed to handle complete engineering tasks from natural-language instructions. It can plan, write, test and verify code across entire repositories, allowing developers to delegate complex software development workflows to AI.
The coding agent uses a multi-agent architecture that coordinates multiple AI agents working in parallel to tackle complex tasks more efficiently. Users can install the tool with a single command, while two pricing tiers are available, including an opt-in contributor tier offered at a significantly lower price point.
Muse Spark 1.2 is the coding-optimised model powering Muse Code. The update was co-trained with Muse Code to provide tighter integration between the model and coding agent, with improvements aimed at increasing first-attempt success, producing cleaner execution and reducing the need for repeated prompts.
The model is designed for long-running, multi-file development workflows and features a 1 million-token context window, enabling it to process and maintain substantially larger amounts of code and project information during development tasks.
Meta said Muse Spark 1.2 is competitive with leading coding models and represents its next step towards frontier AI models, with larger and more capable models expected to follow.
The launch of Muse Code and Muse Spark 1.2 further underscores the growing focus on AI agents capable of carrying out complex, multi-stage software engineering tasks rather than simply generating snippets of code.
The move toward AI agents handling complete coding workflows is impressive. Tools like Muse Code show how development is shifting from code generation to end-to-end execution, while services such as HTMLSave’s HTML-to-URL workflow make it easy to quickly test and share web projects.
ReplyDelete