Meta launches Muse Code: an artificial intelligence assistant that writes code for you
On Wednesday 5 August, Meta —the company that owns Facebook, Instagram and WhatsApp— presented Muse Code, an artificial intelligence tool that helps programmers write, review and fix computer code. It arrives in a trial (beta) version and competes directly with OpenAI’s Codex and Anthropic’s Claude Code. The big difference: Meta is betting on a very low price so that any developer, from a student to a large company, can use it without the bill exploding.
What exactly is Muse Code?
Imagine you have a workmate who knows how to program in every language, never gets tired and can read thousands of files in seconds. Muse Code is something like that: a specialized artificial intelligence agent for programming. It not only suggests individual lines of code (as traditional autocompletes do), but is capable of handling complete tasks: planning a change, writing the necessary code, testing that it works and, if something fails, fixing it on its own.
It is built on Muse Spark 1.2, Meta’s latest artificial intelligence model specifically trained to understand and generate code. The novelty is that Meta has trained the model and the tool together from the start, so that they work like a well-matched couple: the model “thinks” and the tool “executes”.
How it works inside (explained without jargon)
When you ask Muse Code for a big task —for example, “add a payment system to this online shop”—, the tool does not do it all at once in a single thread. Instead, it divides the work into parts and launches several “sub-agents” that work in parallel, each in its own isolated copy of the project. It is as if you hired a team of programmers who share out the tasks and work at the same time without treading on each other’s changes.
Another practical advantage: it saves a history of everything it does. If the computer turns off, the power cuts out or there is a fault, Muse Code can resume exactly where it left off, without having to start from scratch. For anyone who has lost hours of work because of an unexpected shutdown, this is a real relief.
How much does it cost?
Meta has focused on price. The standard plan costs 1.25 dollars per million “input tokens” (what you send it: your code, your instructions) and 4.25 dollars per million output tokens (what it returns to you: the new code, the explanations). To give you an idea: a million tokens is roughly equivalent to 750,000 words in Spanish —much more than fits in a typical project.
In addition, there is a “collaborator” plan at 0.30 dollars per million total tokens, designed for open projects and community contributions. In comparison, OpenAI’s and Anthropic’s alternatives are usually several times more expensive. Meta wants cost not to be a barrier: so that a freelance programmer or a small startup can afford an “AI companion” without thinking twice.
Why does this matter if you are not a programmer?
Programming is at the base of almost everything you use daily: the banking app, the medical appointment website, the system that manages your salary, your console’s video game. Making programming cheaper and faster means the digital tools we all use arrive sooner and better.
In addition, tools like Muse Code lower the barrier to entry. Someone with an idea but without deep technical training can, with the help of AI, build a working prototype. It does not replace the expert programmer —human supervision is still necessary for architecture, security and ethics decisions—, but it multiplies what a person can achieve in the same time.
Meta arrives late to this race (OpenAI and Anthropic have a head start), but it does so with the house’s usual strategy: loosening up on prices and putting the technology within everyone’s reach. If it works, we will see more software, more varied and created by different people. And that, in the end, benefits all of us who live surrounded by code without realizing it.
