Model Spend Arena2ND ED.
builds models · serves endpoints · 2026-08-14

MiniMax

Official site · 🇨🇳 China

Plan

Token Plan (Plus / Max / Ultra)

Tier$/monthIncluded usage
Plus$20.00
Max$50.00
Ultra$120.00

readable by API units unstated

As advertised. no numeric quota published; console shows a usage bar

What that actually means. No token, credit, prompt or dollar figure for any tier -- only '3-4 agents'. Verified directly: the API returns a remaining percentage for text with total_count 0, so even the endpoint hides the denominator. Video quota does return absolute counts.

Where to check what is left. GET /v1/token_plan/remains, plus the console bar

Renamed from "Coding Plan" to "Token Plan" during 2026. THIS IS THE PLAN BEHIND THE sk-cp- KEY USED IN THIS REPO. Third-party token figures conflict badly (1.6B/5.1B/9.8B vs 600M/1.8B/7.1B) and none match official docs, which publish nothing. Only official anchor: a migration guide calling one tier "roughly 12.5B tokens of monthly capacity". Your own console usage bar is more authoritative than anything published. 5-hour + weekly windows; unused quota does not carry over. Overflow credits at 1,000 credits = $1, no markup.

Reverse-engineered. This vendor publishes no figure, so one was derived: about 7.4B tokens a week on Ultra ($120/month), roughly 32B a month, or $0.0038 per million. Derived from a SINGLE account; not an official figure. Robust to the API's integer-percentage resolution (worth about 0.05B) and to the rolling-versus-fixed window mismatch (7.2-7.7B). One caveat that cannot be resolved from outside: the internal weighting of cached input, fresh input and output has not been reverse-engineered, so this figure is valid at the observed ~92% cache-hit rate and should not be extrapolated to a very different one. Note also that the accounting has already changed once: cached tokens formerly did not count against the quota and now do, which raised the effective cost sharply and was not announced.

checked 2026-08-13 · high (price) / low (quota) · source

Models it builds

ModelCoding indextok/s$/M in$/M out$ / sessionCache
MiniMax-M358.670.0$0.30$1.20$0.543494.3%
MiniMax M2.752.60.0$0.30$1.20

Models it serves

3 endpoints, cheapest first. This is the view the model-by-model tables cannot give you: what one provider actually offers, and at what precision.

ModelQuant$/M in$/M outContextUptime 30m
MiniMax-M3fp8$0.30$1.20524,28897.6%
MiniMax M2.7fp8$0.30$1.20204,80097.6%
MiniMax M2.7fp8$0.60$2.40204,800

Where to buy its models

Everyone serving these models, cheapest first, with the numeric precision each one runs. The lab is often not the cheapest place to buy its own model.

ModelServed byQuant$/M in$/M outUptime 30m
MiniMax-M3CoreWeavefp4$0.23$0.9699.9%
MiniMax-M3GMICloudfp8$0.24$0.9697.8%
MiniMax-M3DeepInfrafp8$0.28$1.1098.8%
MiniMax-M3AtlasCloudfp8$0.30$1.2099.7%
MiniMax-M3MiniMaxfp8$0.30$1.2097.6%
MiniMax-M3Morphfp4$0.30$1.2096.7%
MiniMax-M3Novitafp8$0.30$1.2099.7%
MiniMax-M3Parasailfp8$0.30$1.2099.0%
MiniMax-M3StreamLakefp8$0.30$1.2099.5%
MiniMax-M3Togetherunknown$0.30$1.2099.5%
MiniMax-M3Venicefp8$0.30$1.2099.8%
MiniMax-M3ModelRunfp4$0.75$3.0099.9%
MiniMax M2.7Maraunknown$0.24$0.96100.0%
MiniMax M2.7DeepInfrafp8$0.25$1.00100.0%
MiniMax M2.7GMICloudfp8$0.27$1.08100.0%
MiniMax M2.7Novitafp8$0.27$1.0898.2%
MiniMax M2.7AtlasCloudfp8$0.30$1.20
MiniMax M2.7Fireworksunknown$0.30$1.20
MiniMax M2.7MiniMaxfp8$0.30$1.2097.6%
MiniMax M2.7DeepInfrafp8$0.38$1.7099.5%
MiniMax M2.7Groqunknown$0.60$1.80100.0%
MiniMax M2.7MiniMaxfp8$0.60$2.40
MiniMax M2.7SambaNovaunknown$0.60$2.40
◆ Some links on this page are referral links: if you sign up through them this site may earn a commission, at no extra cost to you. Rankings, prices and measurements are taken from the sources listed under Method, and are not influenced by whether a provider has a referral programme.