Written By Deepti Ratnam
Written By Deepti Ratnam
Edited by Deepti Ratnam |Published: Aug 06, 2026, 10:22 AM (IST)
Meta expands AI developer tools with Muse Code and Muse Spark 1.2
Meta has announced the rollout of a new AI coding agent called Muse Code (beta), which is equipped with its latest AI model, Muse Spark 1.2. The company claims the new tools will enable developers to write code, debug, comprehend large code bases and finish software projects quicker.
According to Meta, Muse Code is its next step towards building more powerful AI models for software engineering. Even more powerful AI models are being developed by the company.
Muse Code is an AI assistant for the terminal. It has been designed to enable developers to efficiently manage their complex programming workloads in large-scale software projects.
Muse Code doesn’t just generate code, it can:
The background AI agents are one of the largest features of Muse Code.
These remain with the main AI agent during the coding process. They can collect information, perform less intensive tasks and feedback when necessary.
These agents remain active, which minimizes delays and allows the AI to finish long and complex tasks more quickly.
During each coding session, Muse Code maintains a detailed event log, according to Meta.
Every AI response, tool action, code edit and approval is logged. The system can be restarted from the last point it left off, if there is a crash or an unexpected stop.
This makes Muse Code appropriate to coding tasks that can last hours or even a whole day.
Muse Code contains a number of pre-installed commands to make software development easier.
These include some of the following:
The tools are created to supply a structured and error discounting approach to coding.
Muse Spark 1.2 is Meta’s newest AI model for coding. It replaces Muse Spark 1.1 and offers some enhancements.
Meta claims the new model excels in the following areas:
The company says it used more computing power and trained the model on a wider variety of coding environments to improve its performance.
According to Meta, Muse Spark 1.2 has been trained on long and complex programming tasks.
These include:
When working on large projects, the model plans the work, monitors progress and recalls key information.
So, it’s interesting that Meta used Muse Spark 1.1 to train Muse Spark 1.2.
The older model was used to create a coding problem and to determine if the new model solved it properly.
This ongoing training process has enabled Muse Spark 1.2 to learn to better follow complex instructions and solve complex coding problems, Meta said.
Meta also released a real-life test to illustrate the model’s capabilities.
The company requested Muse Spark 1.2 to streamline GPUs kernels running on NVIDIA Hopper graphics cards.
The AI performed over 1,000 tool calls in the test period, up to 24 hours. It was always writing code, compiling code, testing, and optimizing the code.
Meta states that the AI has garnered a lot of performance enhancements over the original implementation.
Starting today, Muse Spark 1.2 is available via Muse Code and the Meta Model API.
In the future, the company will roll out additional functionality for Muse Code, as well as even stronger AI models, the company states. Meta is continuing to develop new AI features for developers with Muse Code and Muse Spark 1.2 and is looking to help software developers make the process faster, simpler and more efficient.
