Meta's Contributor tier: tokens up to 21× cheaper, paid for with your data

Meta's Muse Spark 1.3, released 2 September 2026, is priced at $1.25 per million input tokens and $4.25 output — unchanged from 1.2, with the same 1M-token context. That is not the interesting part.

The interesting part is a second price for the same model: $0.10 input and $0.20 output, with cached input at $0.002. Meta calls it the Contributor tier. It is roughly 12 times cheaper on input and 21 times cheaper on output, and the difference is not a discount. It is a purchase — paid in a different currency.

What the Contributor tier actually costs

The terms, as Crypto Briefing summarises them: on the Contributor tier, "your prompts, responses, and usage patterns get fed back into Meta's training pipeline." Standard-tier traffic is not. That is the entire difference between the two prices.

Put plainly, the model is the same; what changes is who owns what you do with it. On Standard, a company's prompts and outputs are its own. On Contributor, they become training data for the next Muse — including, for a coding workload, the code being written and the codebase context it is written against.

Why this is a genuinely new pricing shape

Vendors have offered opt-in data sharing before, usually as a checkbox with a vague benefit. What Meta has done is put a number on it — a large one — and made it a tier you select rather than a setting you toggle. At a 21× spread on output, a team spending $10,000 a month on Standard could spend under $500 on Contributor for the same tokens. That is the kind of gap that gets noticed by a finance function that never reads a data-usage clause.

It is also a mechanism the rest of the market will study. Meta's stated reason for the tier is model improvement; the training signal from real production usage is worth more to a lab than the token revenue it forgoes. If the uptake is meaningful — and this week's reporting suggests Meta is pleased with it — expect other labs to offer the same trade.

Who should and should not take it

Reasonable candidates: experimentation, learning, prototypes, internal tooling on non-sensitive data, open-source work, anything a team would be comfortable publishing anyway. For those workloads, the 12–21× saving is real and the cost is close to nothing.

Not candidates, regardless of price: anything touching customer data, regulated data, proprietary code a company considers a competitive asset, contractual work with confidentiality terms, or any workload where a client has not consented to their material entering a third party's training set. A 21× discount does not cover a data-protection breach or a client lawsuit.

The grey middle, which is most of real usage: a product team's own code, internal documents, support conversations. Here the honest question is not "is it allowed" but "would we be comfortable if a competitor's model was noticeably better at our domain next year because of what we sent." For some teams that is fine. For others it is the whole game.

The practical test

Before choosing Contributor for any workload, a team should be able to answer three questions in writing: what data will pass through it; whether any of that data belongs to someone else; and whether anyone who owns that data has agreed to this use. If any answer is "not sure," the tier is not cheap — it is unpriced.

TrustList records Muse Spark 1.3 on the AI-model board at its Standard price with the Contributor tier described in the listing rather than as a separate model. Same weights, different terms; the listing should say so rather than show a $0.10 headline that most business buyers cannot actually take.

---

Sources

• Crypto Briefing — Meta offers discounted access to Muse Spark 1.3 for data sharing (pricing for both tiers, data-usage terms)
• OpenRouter — Muse Spark 1.3 Contributor (Contributor-tier listing and price)
• Axios — Meta debuts Muse Spark 1.3 as personal agent work continues (release; the article could not be retrieved in full at research time, so no uptake figure is quoted here)