Meta practically gives away new AI models — 29% cheaper than DeepSeek

Meta is practically giving away its newest models — provided you're willing to help train the next ones.

The company has introduced a Contributor tier for Muse Spark 1.2 and Muse Code. Agree to let Meta use your prompts and model outputs to train subsequent models, and you get a steep discount on API pricing.

At that tier, input, cache-hit, and output tokens cost $0.10, $0.002, and $0.20 per million tokens, respectively. DeepSeek V4 Flash, by comparison, charges $0.14, $0.0028, and $0.28 per million tokens for the same categories — which makes Muse about 29% cheaper across the board.

Not willing to share your data? Then you'll be on the standard tier.