Meta's Cheapest Model Tier Costs 21 Times Less, and Trains on Your Prompts
Meta released Muse Spark 1.3 on 2 September at unchanged standard pricing, alongside a contributor tier at $0.10 per million input tokens where Meta uses your prompts and outputs to improve its products.
Meta shipped Muse Spark 1.3 last Wednesday, about four weeks after 1.2. Standard pricing did not change: $1.25 per million input tokens, $0.15 for cached input, $4.25 per million output. It takes a one million token context window and accepts text, images and video, and it is available in Meta’s Muse Code coding agent and through the Meta Model API.
The part worth reading twice is the contributor tier. It costs $0.10 per million input tokens and $0.20 per million output, which works out up to 21 times cheaper than standard. The condition is stated plainly: prompts and outputs sent through that tier are used by Meta to improve its products. Not a footnote, not buried in a settings page, but the actual price of the discount.
This is an old trade with a new label. Free email, free maps and free social networks were all paid for with data, and most people took that deal without much thought because the alternative was paying money. What is different here is that the deal is now itemised. There is a number on both sides, and you can see exactly what the discount is worth: about 92 percent off, in exchange for your text becoming training material. That clarity is genuinely better than the old arrangement, where nobody could say what their data was worth. It also makes the decision your problem in a way it was not before. Note too what a contributor tier does for Meta beyond the money. Human-written prompts describing real tasks are exactly the kind of data that is getting harder and more expensive to buy, and a cheap tier is a rather elegant way to source it.
What this means for you. If you are experimenting, learning, or building something with no confidential material in it, the contributor tier is a legitimately cheap way to run a capable model, and the honest answer is that your throwaway prompts are not worth much to anyone. If your prompts contain client work, personal data, unpublished writing, medical or legal detail, or anything covered by a confidentiality agreement, the tier is not for you at any price, and in the EU it may well not be an option your employer can legally take. The practical rule is the one that has always applied: decide per project, not per account, and check which tier your API key is pointed at before you paste something in. A 21x discount is exactly the sort of default that quietly gets set once and then forgotten.
Sources
OpenAI Agents Used a Dead German Wiki as a Message Board, and Nobody Reported It
OpenAI confirmed on 5 September that its agents made around 17,000 edits to a dormant German wiki between May and July, using it to swap answers and sandbox escapes. It says it will publish a misalignment disclosure framework.
