minimax API Pricing

Short answer. MiniMax runs a permanent 50% discount against its own list price, tiers by context length with a hard doubling above 512K tokens, and sells a priority tier at 1.5x. The headline $0.30/$1.20 is real, but it applies only below the context cliff.

The structure

Model	Input	Output	Out/in	Cache	Condition
MiniMax-M3	$0.30	$1.20	4.00	$0.06 (20%)	≤512K input
MiniMax-M3 (>512K)	$0.60	$2.40	4.00	$0.12	>512K input
MiniMax-M2.7	$0.30	$1.20	4.00	$0.06	≤512K
M2.7 Highspeed	$0.60	$2.40	4.00	$0.06	faster serving
Priority tier	1.5x	1.5x	—	—	service_tier=priority

The price history recorded here shows M3 cut from $0.60/$2.40 to $0.30/$1.20 — in other words, the “permanent 50% discount” is a halving that has been left in place, not a promotional rate.

The 512K cliff

This is the single most important fact on the page. A request with 600K input tokens is not charged a blended rate; the whole call moves to the higher tier. For long-document workloads the effective price doubles the moment a document crosses the threshold, and a 1M-context model whose price doubles at half its window is a different proposition from one priced flat.

Architecturally, that argues for chunking above 512K, or for comparing the effective price at your real input size rather than at the headline.

Priority at 1.5x is cheap insurance

Most providers price priority implicitly or not at all. A 50% premium for a guaranteed service tier is modest, and for customer-facing workloads it is usually the cheapest reliability purchase available.

Our view

The context cliff is the point of this page. A flat-looking $0.30 input that doubles above 512K is a worse deal for long-context work than a provider charging $0.45 flat. Compare at your actual context length, not at the marketing number.

A permanent 50% discount means the list price is fiction. MiniMax’s list price is not a price anyone pays; it exists to make the discount legible. Treat the discounted number as the real one when comparing across providers.

Priority at 1.5x is the most honest reliability pricing in the database. It is stated, bounded and optional — which is more than most competitors offer, where priority is bundled or unstated.

M3 and M2.7 at identical prices is a migration waiting to happen. M2.7 has a 204K context and a non-commercial licence; M3 has 1M and a commercial one. There is no price reason to stay on M2.7.