What happened
On August 5, Meta released Muse Spark 1.2, a coding-focused update to its AI model line, alongside Muse Code, a tool built for long-horizon agentic tasks, the kind where an AI works through a large project step by step rather than answering one question. According to analysis by researcher Simon Willison, the model was trained with heavy emphasis on coding work, including whole-repository generation and large end-to-end projects, and was co-trained with the Muse Code tooling so the two perform best together. The pricing is the notable part. The standard tier costs 1.25 dollars per million input tokens and 4.25 dollars per million output tokens, in line with comparable models. But a contributor tier drops that to 10 and 20 cents respectively, a discount of roughly 85 to 92 percent, in exchange for consenting to let Meta use your data to improve its products.
Why it matters for your business
The discount-for-data trade is becoming a standard pattern in AI pricing, and it is exactly the kind of fine print that catches small businesses. If you hire a developer or agency who uses AI coding tools, or your own staff does, the cheap tier means the prompts and code sent to the model can feed the vendor's training. That is often fine for boilerplate, and a real problem if the material includes client information, credentials, or business logic you are contractually obliged to keep confidential. The practical move is not to avoid these tools, which are genuinely getting cheaper and more capable. It is to know which tier is in use. Ask your developer or agency one direct question: does the AI tooling you use on our project train on our data? If the answer is yes or unknown, ask for the standard tier or an agreement that excludes your material.