According to CoinMeta, Meta has announced that its new model is almost free of charge, with a price 29% cheaper than that of DeepSeek. Meta has set low-price tiers for Muse Spark and Muse Code; users are allowed to use prompt and the model output to train subsequent models, thereby receiving significant discounts. The prices for this tier per million token of input, cache hits, and output are $0.10, $0.002, and $0.20 respectively, while the prices for DeepSeek, v4, and flash are $0.14, $0.0028, and $0.28 respectively. If users do not wish to share their data, they will need to use the standard tier, with input, cache hit, and output prices rising to $1.25, $0.15, and $4.25 respectively, which is 12.5 times, 75 times, and 21.25 times the low-price tier prices. The contributor tier allows for a maximum of 60 requests per minute and 2.1 million token, whereas the standard tier supports 3,000 requests per minute and 4 million token. DeepSeek has just announced that API will see a significant price increase soon, but the specific prices have not yet been disclosed.