Contributor vs. Standard: doing the tier math
The two ways to pay for Muse Spark 1.2 differ by more than an order of magnitude. The right choice isn’t one tier — it’s a per-project policy.
The numbers
Muse Code itself is free to install; you pay for Muse Spark tokens through the Meta Model API. Two model variants, two price books. muse-spark-1.2-contributor: $0.10 per million input tokens, $0.002 cached input, $0.20 output. muse-spark-1.2 (standard): $1.25 input, $0.15 cached input, $4.25 output. Same model capability, same 1M-token context window — a 12.5x–21x price gap.
What you’re actually trading
The gap is priced data. On the contributor tier, your usage may be used to improve Meta’s products — that’s the deal, stated plainly. On standard, it never is. And for organizations where even standard retention is too much, Meta has begun accepting requests for zero data retention through Meta sales — a signal that enterprise procurement is being courted, not deflected.
Notice what the gap is not: a capability tier. You aren’t buying a smarter model with standard — you’re buying a privacy posture. That’s unusual in a market where cheap tiers typically mean smaller models, and it makes the decision cleanly one-dimensional: how sensitive is this code?
A per-project policy beats a blanket choice
Because /model switches the backing model mid-session, the tier decision doesn’t have to be made once. A sensible default policy: open-source work, side projects, and learning — contributor, where a heavy week of agentic coding costs less than a coffee; anything proprietary, client-owned, or regulated — standard, with zero-data-retention paperwork for the strictest cases. Usage bills at whichever model served the tokens, so mixing within a day is fine.
The cache line matters more than it looks
Agentic coding is unusually cache-friendly: the repository context that dominates input gets re-sent turn after turn, and cached input bills at a fraction of fresh input — $0.002 versus $0.10 on contributor, $0.15 versus $1.25 on standard. Long sessions in one repository are therefore much cheaper than the headline input price suggests. Add /effort — thinking tokens bill as output, the expensive column — and the two commands together are most of practical cost control. The pricing page has an interactive estimator if you want to plug in your own volumes.
The competitive read
Standard-tier pricing lands in the same range as rival frontier coding APIs — the contributor tier is the aggressive move, an order of magnitude below the market for developers willing to share data. It’s a classic scale play: Meta gets training signal from real-world agent sessions; price-sensitive developers get a frontier agent at hobbyist rates. Both sides of that trade should go in with eyes open — which is exactly why it’s good that the trade is explicit.