Meta Put a Price on Your Data: the Muse Spark Contributor Tier Is 92% Off If It Can Train on Your Prompts
- What happened
- Meta's Model API docs now publish two price lists for the same Muse Spark models: Standard at $1.25/$4.25 per 1M tokens with no training on your data, and Contributor at $0.10/$0.20 in exchange for training rights on your prompts and completions.
- Why it matters
- It is the clearest public price a frontier vendor has put on customer data, and the discount hides a 30x rate-limit cut: Contributor gets 100 requests per minute against Standard's 3,000.
- What to do
- Check your request rate before your budget. If you clear 100 RPM, or your prompts carry anything confidential, the Contributor discount is not available to you in practice.
The verdict
Take the Contributor tier for prototyping and throwaway experiments. Keep production on Standard. The deciding factor is not the privacy trade, it is the rate limit Meta printed underneath the discount.
What happened
Meta's Model API docs now carry two parallel rate cards for the same models. Standard covers muse-spark-1.3, muse-spark-1.2 and muse-spark-1.1. Contributor covers muse-spark-1.3-contributor and muse-spark-1.2-contributor, at roughly a tenth of the price, in return for training rights on what you send. Our entries: Muse Spark 1.3, Muse Spark 1.2, Muse Spark 1.1.
| Tier | Models | Input / 1M | Output / 1M | Cached input |
|---|---|---|---|---|
| Standard | muse-spark-1.3, 1.2, 1.1 | $1.25 | $4.25 | $0.15 |
| Contributor | muse-spark-1.3-contributor, 1.2-contributor | $0.10 | $0.20 | $0.002 |
Meta's wording on Standard: "your prompts and completions are not used to train Meta models." Contributor offers "heavily discounted token pricing in exchange for permission to use your prompts and completions to train future Meta models," which Meta frames as lowering "the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable."
Where the 92% comes from
Meta publishes two rate cards, not a discount percentage. The figure is derived from its own numbers: input drops from $1.25 to $0.10 per 1M tokens, a 92% cut. Output drops from $4.25 to $0.20, a 95% cut. Cached input drops from $0.15 to $0.002, a 98.7% cut. TechCrunch, covering the same rate card, rounds the package to "about 95%." We lead on the input number because input is what a coding agent burns.
The catch under the discount
Rate limits invert between the tiers.
| Tier | Requests per minute | Tokens per minute |
|---|---|---|
| Standard | 3,000 | 4M |
| Contributor | 100 | 3M |
A 30x cut in request rate, and the limits apply per team, not per API key. Contributor is priced for volume and throttled against it: you can burn a lot of tokens slowly, or a few tokens fast, and the cheap tier only offers the first. For a coding agent firing many small requests, 100 RPM is a wall you hit long before price becomes the deciding factor.
A second catch sits outside both rate cards. Artificial Analysis puts Muse Spark 1.2 at $0.40 per task at a score of 57, and estimates $0.55 per Intelligence Index task for the shipping 1.3 xhigh configuration at a score of 61: about 37% more to finish the same task, while Meta kept Standard pricing exactly where it was at $1.25 / $4.25. Artificial Analysis attributes the increase primarily to heavier input-token consumption on agentic evaluations, as reported by VentureBeat(opens in new tab). The rate card is not the price. An unchanged per-token rate can still bill you more per unit of work done.
Contributor is also not offered on Muse Voice Transcribe at launch. Other rates confirmed on the same page: no long-context premium on any tier ("you pay the same rate whether your context window is mostly empty or almost full"), web search grounding at $2.50 per 1,000 queries, Muse Image at a flat $0.01 per generated image, and Muse Voice Transcribe at $0.18 per hour of audio.
Why it matters
TechCrunch reported on September 3 that this class of user data is "vital for making agentic tools work better," and that Meta's own initiative to track the computer usage of its employees, launched earlier this year, "attracted wide internal criticism and was paused in June." Buying the data at the API boundary is the cleaner route to the same input.
Princeton's Arvind Narayanan, quoted in the same piece, points out how much companies already pay to keep their data out: firms "stick with token-billed Enterprise plans even though the subscription-based consumer plans like Claude Max and ChatGPT Pro are discounted by 10x-20x or even more."
What is new here is the framing. Most vendors bury training rights in terms of service and publish one price. Meta printed two numbers side by side and let the gap name what your data is worth to it: about $1.15 per million input tokens. An explicit price beats a buried clause, even when the price is low.
The trade also looks durable rather than promotional. Standard rates have not moved across 1.1, 1.2 and 1.3, so the discount is a standing structure attached to a model generation, not an intro offer that expires when the next release lands.
What changes for you
The privacy question is the easy one. If your prompts carry customer data, proprietary code, or anything under a confidentiality obligation, Contributor is not a choice you get to make. Stay on Standard.
The harder one is that many teams will read a 92% discount, skip the rate-limit table, and meet 100 RPM in production. Check your request rate first, then your data policy, then the price.
Directory status
No directory rating moves on this. Standard-tier pricing is unchanged, and Muse Spark 1.3 is now listed at the same $1.25 / $4.25 as 1.2 and 1.1, holding its conditional verdict.
What to do
- 1 Measure your peak requests per minute before you look at the price. Contributor caps at 100 RPM against Standard's 3,000, per team rather than per API key, and an agent workload hits that ceiling long before the discount pays off.
- 2 Audit what your prompts actually contain. Customer data, proprietary code, or anything under a confidentiality obligation rules out Contributor regardless of price.
- 3 If you move prototyping traffic to Contributor, pin the model id explicitly (muse-spark-1.3-contributor) and keep production on the standard id. The two are separate model ids, not a billing flag.
- 4 Do not plan on Contributor for speech work: the discounted tier is not offered on Muse Voice Transcribe at launch, which stays at $0.18 per hour of audio.
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.