Meta launched its first coding agent, Muse Code, in preview, running on Muse Spark 1.2, the model Meta trained alongside the agent to improve its coding performance. Developers set it up with a single command and hand it whole engineering tasks, from planning a change through writing the code and checking the result, AI chief Alexandr Wang told CNBC. Pay-as-you-go pricing matches the Muse Spark 1.1 API at $1.25 per million input tokens and $4.25 per million output, and Wang said a contributor tier costs more than 10 times less for developers who let Meta use their data to train the model. Meta has also begun taking zero-data-retention requests, which keeps enterprise data out of training entirely. Wang said Meta is competing with Anthropic and OpenAI on price rather than capability.
