SAEDNEWS: Meta has entered the increasingly competitive AI coding market with Muse Code, a beta terminal-based coding agent powered by its new Muse Spark 1.2 model. Beyond software-engineering capabilities, Meta is putting an unusually strong emphasis on lower usage costs as it takes aim at products from OpenAI and Anthropic.
According to SaedNews: Meta is making a new move in one of the most closely watched corners of artificial intelligence: AI-powered software development. The company has unveiled the initial beta of Muse Code, a terminal-based coding agent designed to compete with products such as Claude Code from Anthropic and Codex from OpenAI.
The launch is significant not simply because Meta is releasing another AI tool, but because Muse Code is aimed at a task that has quickly become central to the AI industry: allowing developers to describe what they want in ordinary language and have an AI system help turn those instructions into working software.
At the heart of Muse Code is Muse Spark 1.2, a new version of Meta's AI model. According to the company, the updated model brings improvements in code generation, complex debugging, understanding large codebases and handling end-to-end developer workflows.
That means Muse Code is intended to go beyond simply suggesting lines of code.
The agent can reportedly write code, plan changes, validate results and help carry out broader software-engineering tasks. It can also coordinate multiple sub-agents, effectively distributing different parts of a task to other agents in an effort to complete complex jobs more efficiently.
That capability could become particularly important as AI coding tools move from simple assistants toward systems capable of managing larger portions of a development process.
Meta also used several demonstrations to show what the new system can produce. Its examples included an interactive model of a photon sphere, a copy of the game Plants vs. Zombies, and a demonstration of a tool that creates a webpage based on an MP4 file.
But the most commercially interesting part of the announcement may not be the demonstrations at all.
It could be the price.
Meta appears to be positioning affordability as one of Muse Code's major advantages. The company says the tool will use the same pay-as-you-go pricing as Muse Spark by default: $1.25 per million input tokens and $4.25 per million output tokens.
Meta's chief AI officer, Alexander Wang, also told CNBC that the company plans to offer a contributor tier designed to give users access at a substantially lower cost.
According to The Wall Street Journal, users who choose that tier would agree to provide feedback intended to help improve the coding agent. The reported price would be just $0.10 per million input tokens and $0.20 per million output tokens.
That is considerably below the cited pricing for Anthropic's Sonnet 5 model, which is typically listed at $3 per million input tokens and $15 per million output tokens.
The difference could matter enormously for developers and businesses that use AI coding systems heavily. For occasional users, token pricing may seem like a technical detail. For organizations running large numbers of AI-assisted development tasks, however, those costs can accumulate quickly.
That gives Meta a straightforward way to make Muse Code more attractive: compete not only on what the model can do, but also on what it costs to use.
The strategy has echoes of a broader trend in the AI industry. Chinese AI companies have already attracted attention by offering lower-cost alternatives, with some businesses turning to models such as DeepSeek as they look for ways to reduce the rising expense associated with American AI services.
Muse Code therefore arrives at a moment when the competition is no longer limited to who has the most capable model. Price, workflow integration and the ability to handle complicated tasks are becoming equally important parts of the battle.
For developers, the appeal of an AI coding agent is increasingly tied to how much work it can take off their hands. Writing code is only one part of software development. Planning modifications, understanding an existing codebase, testing changes and checking whether the final result works can consume considerable time.
Meta's approach is clearly aimed at bringing those steps together inside one agent-driven workflow.
Still, the launch is a beta release, and the real test will come from how reliably Muse Code performs outside demonstrations. A coding agent may look impressive when completing carefully selected examples, but developers ultimately need systems that can handle messy, evolving projects without introducing new problems.
The pricing, meanwhile, gives Meta another potential advantage. If the system delivers useful results while maintaining its lower cost structure, it could become an appealing alternative for users who are increasingly conscious of AI expenses.
The arrival of Muse Code adds another major name to an already crowded AI coding market. With Meta now challenging established offerings from OpenAI and Anthropic, developers may soon have even more choices over which AI systems they trust with their software projects.
And that could make the next phase of the AI coding race less about simply having the smartest assistant—and more about which company can offer the best combination of capability, reliability and cost.
