Meta Launches Muse Code But Trails Anthropic on Benchmarks

Meta has officially released Muse Code in beta, marking the company's first major entry into the AI coding agent space. Powered by the new Muse Spark 1.2 model, the tool handles software engineering tasks like planning changes and writing code. However, benchmark charts published by Meta show the tool losing to competing models across multiple tests.

On the TerminalBench test, Muse Spark 1.2 scored 82.9 percent, falling behind Anthropic and other top systems. Independent data suggests the performance gap is even wider because Meta left the strongest rival models off its official comparison charts. Despite finishing behind the competition in raw accuracy, Meta is positioning the product as a more affordable option for developers.

The pricing strategy comes as Meta faces rising capital expenditures and shrinking operating margins. The company is offering the new coding agent at a steep discount compared to market leaders, while also working on background agents that offer unique workflow features.

Developers now face a choice between top tier accuracy or significant cost savings. The coming weeks will show if lower prices can attract users away from market leaders.