As recorded at collection: Muse Spark 1.2 / Muse Code, prices up to 21 times lower

Meta's coding models Muse Spark 1.2 and Muse Code have a low-cost Contributor plan. The price per 1 million tokens is 0.10 dollars for input, 0.002 dollars for cache hits (when stored content is reused), and 0.20 dollars for output. The corresponding prices for competitor model DeepSeek V4 Flash are 0.14 dollars, 0.0028 dollars, and 0.28 dollars, making Contributor about 29% cheaper across all three. DeepSeek has announced plans to raise its API prices significantly, but has not yet disclosed the specific increase.

To use the Contributor plan, you must allow Meta to use your prompts and the model's responses to train future models. If you do not want to provide data, you must use the Standard plan. Standard costs 1.25 dollars per 1 million input tokens, 0.15 dollars per 1 million cache-hit tokens, and 4.25 dollars per 1 million output tokens, which is 12.5 times, 75 times, and 21.25 times the Contributor plan prices, respectively.

The Contributor plan is limited to 60 requests and 2.1 million tokens per minute. These limits are lower than Standard's 3000 requests and 4 million tokens per minute, so tasks that process large volumes in a short time may hit the limits. Prompts and responses are used for model training, so the plan is not suitable for entering sensitive information.