Meta Joins the Coding-Agent War With a Cheaper Muse Code Bet

Meta Joins the Coding-Agent War With a Cheaper Muse Code Bet
Meta is stepping into one of AI’s fiercest contests: turning models from code assistants into agents that can take responsibility for software work. Its wager is Muse Code, a beta product that aims to make large-scale engineering cheaper without making it hands-off.

The company launched Muse Code Wednesday, pairing the terminal-based agent with Muse Spark 1.2, a coding-focused model update. Meta says the tool can tackle “complex software engineering tasks across large repositories: planning changes, writing code, and validating the results.” That puts it beyond autocomplete: the promise is an agent able to inspect a codebase, implement a change and test what it has done.

Meta also offered a visible early demonstration. Chief executive Mark Zuckerberg showcased a game built by the agent and made it available through the company’s research blog, according to one report. Yann LeCun amplified the launch message, describing Muse Code as a beta release for “complete software engineering tasks across large repos.”

The timing is pointed. Anthropic’s Claude Code won early developer attention, while OpenAI has expanded Codex through ChatGPT and developer terminals. Meta had previously signalled that developer tools were not its central focus; Muse Code now reads as a direct bid for the same programmers.

Its sharpest competitive argument is price. Meta AI chief Alexandr Wang called Muse Code “one of the most affordable coding agents on the market, available globally.” Muse Spark 1.2’s standard pricing starts at $1.25 per million input tokens, while a discounted contributor tier lets Meta use activity to improve its products. That split may suit solo developers and startups, but companies guarding proprietary code may see the standard tier as the safer choice.

Continue reading https://foxvector.com/stories/019fd4d0-7703-163c-72fc-140eb6b146ea