MiniMax API Pricing (2026)

MiniMax-M2, the default text tier, costs $0.255 per million input tokens and $1.00 per million output via OpenRouter, against $0.30/$1.20 on MiniMax's own platform, while the newer M3 carries a permanent 50%-off promotional rate. That spread between the routed and official price is the first thing most buyers miss, and it is one of three decisions this guide untangles: which tier, which route, and what your real workload bills. We reached every MiniMax tier through OpenRouter and timed the calls ourselves, because MiniMax has no native US billing key yet and the routed number is what you actually pay.

MiniMax is the Shanghai lab behind the Hailuo models, and its text API splits into three tiers that do not map cleanly onto "cheap, mid, premium." M2 is the workhorse, M1 the legacy model that still bills at a premium, and M3 the new release whose headline price is a standing discount.

MiniMax API pricing cluster (verified 2026-06)

According to MiniMax, the platform lists each text tier separately, and the per-token rates diverge enough that picking the wrong tier can double your bill. Each article below answers one buyer question with a verified price and, where we could measure it, a first-hand timed call.

Sub-topic	Article	The number that matters
The default M2 tier rate	MiniMax-M2 API pricing	$0.255/$1.00 routed, $0.30/$1.20 official
The legacy M1 tier	MiniMax-M1 API pricing	$0.40/$2.20 routed; output runs premium
MiniMax vs DeepSeek cost	MiniMax vs DeepSeek cost	DeepSeek V4-Flash $0.14/$0.28 undercuts M2
MiniMax vs GPT-4o cost	MiniMax vs GPT-4o cost	GPT-4o $2.50/$10 is ~10x M2 output
The cheapest MiniMax tier	Cheapest MiniMax model	minimax-01 at $0.20/$1.10 routed
Coding workload cost	MiniMax coding cost	M2 coding call 38/200 tokens in 2.95s
US and international access	Is MiniMax available in the US?	reachable via platform.minimax.io
Budgeting your own spend	MiniMax API cost calculator	blended cost depends on output share

The routed prices above were read from the live OpenRouter catalog on 2026-06-26. The official MiniMax figures are sourced research carrying a native-key re-verification flag, because we could not yet confirm them against a first-party invoice.

What we found by calling the tiers, not just reading the rate card

Documentation gives you a list price. A live call tells you what the route bills and how fast each tier answers, and three points sharpen once you measure them.

First, the routed price genuinely beats the official one on M2. The OpenRouter catalog lists minimax-m2 at $0.255 input and $1.00 output, a touch under the $0.30/$1.20 MiniMax publishes on its own platform. For a US developer with no native MiniMax billing relationship, the route is the only path and also the cheaper one.

Second, the tiers are not priced in a tidy ladder. M1, the older model, bills at $0.40/$2.20 routed, more expensive on both sides than the newer M2 at $0.255/$1.00. Buyers who assume the legacy model must be the budget option get this backwards. If you are on M1 out of inertia, M2 is both newer and cheaper.

Third, latency tracks the workload, not the tier name. A short coding prompt to M2 through OpenRouter came back in 2.95 seconds; a general prompt to the same model took 5.08 seconds, and M3 landed at 5.20 seconds. The faster coding turn reflects a shorter generation.

We timed an M2 coding call (first-hand evidence)

We wanted to know what a real coding request bills, so we called minimax-m2 through OpenRouter with a small code-generation prompt and recorded the usage object the API returned.

Tier (via OpenRouter)	Prompt tokens	Completion tokens	Billed cost	Latency
minimax-m2 (coding)	38	200	$0.0002514	2.95s
minimax-m2 (general)	39	200	$0.000209945	5.08s
minimax-m3 (general)	195	200	$0.00027114	5.20s

The coding call used 38 input tokens, generated 200 output tokens (it hit our 200-token output cap), billed $0.0002514, and returned in 2.95 seconds, measured via OpenRouter rather than a native MiniMax key. The cost shape is the lesson: at $1.00 per million output against $0.255 input, the 200 output tokens drove most of that fraction-of-a-cent charge. Output is where MiniMax bills you, so output-heavy work like code generation or long-form drafting is where the tier choice bites hardest.

The decision matrix: which tier and which route

No single sub-article answers the question buyers actually arrive with: "given my situation, which tier and route do I pick?" Here is the cross-cutting view that ties the cluster together.

Your situation	Pick this tier + route	Where to read
US developer, general workload, no native key	M2 via OpenRouter at $0.255/$1.00	MiniMax-M2 pricing
Absolute floor on input cost	minimax-01 at $0.20/$1.10 routed, accept older quality	Cheapest MiniMax model
Output-heavy coding or generation	M2, and budget against the $1.00 output rate	MiniMax coding cost
Evaluating MiniMax's newest model	M3 at the permanent 50%-off promotional rate	below, then the calculator
Already on legacy M1 out of habit	Move to M2; it is newer and cheaper on both sides	MiniMax-M1 pricing
Deciding MiniMax vs the alternatives	Compare against DeepSeek and GPT-4o before committing	MiniMax vs DeepSeek

The ordering matters. For most US developers the tier decision is nearly made for them: M2 via the route is the cheapest sensible default, and the only reasons to look elsewhere are a hard input-cost floor (which sends you to minimax-01) or a wish to trial the newest model (which sends you to M3). The route is not really a decision right now, because without a native MiniMax billing key the routed path is the path, which is why every price here is the routed value, disclosed as such.

The M3 promotional rate is structural, not a sale

One MiniMax-specific fact is easy to misread as a countdown timer. According to MiniMax, the MiniMax-M3 promotional rate is a permanent 50%-off the raw $0.60/$2.40 card for inputs up to 512K, landing the effective price at $0.30/$1.20. Permanent promotional pricing is a standing discount baked into the published rate rather than a time-boxed sale, so unlike a launch promo it does not expire. According to MiniMax, the above-512K input tier is access-limited at the full $0.60/$2.40, so the self-serve discount applies to the up-to-512K band.

We will not quote a specific context window for the legacy M1 tier, because the sources we checked disagree and we would rather say "verify it in the console" than print a number we cannot stand behind. Context limits change quietly between model revisions, so confirm it against your own account before you architect around it.

How MiniMax sits against the alternatives

MiniMax is mid-pack on price, and two reference points make the tier choice concrete. According to DeepSeek, DeepSeek's V4-Flash bills $0.14 input and $0.28 output on its official endpoint, undercutting MiniMax-M2's routed $0.255/$1.00 on both sides, with the gap widest on output where DeepSeek is roughly a quarter of M2's rate. We anchored that with a first-hand DeepSeek call: a coding prompt of 27 input and 86 output tokens returned in 1.96 seconds, faster than our routed MiniMax coding turn, though the asymmetry matters because DeepSeek was measured natively and MiniMax through OpenRouter.

At the other end, GPT-4o lists at $2.50/$10 and Claude Sonnet 4.6 at $3/$15, putting MiniMax-M2's $1.00 output at roughly a tenth of GPT-4o's and a fifteenth of Claude's. The verdict: MiniMax is not the cheapest Chinese option (DeepSeek is, on raw rate) but it sits far below the Western frontier models, so the real question is rarely "MiniMax or GPT-4o" on price alone, it is "DeepSeek or MiniMax" once you have decided a Chinese-lab model fits the task. The two head-to-head micros below work through that call.

Methodology

MiniMax prices here were read from the live OpenRouter catalog on 2026-06-26 and confirmed against timed API calls routed through OpenRouter the same day; the official MiniMax platform figures are sourced research pending first-party-invoice re-verification, which is why this page carries a native-reverify flag. The DeepSeek anchor was measured on its official api.deepseek.com endpoint, so the MiniMax-vs-DeepSeek latency comparison spans two routes and we disclose that rather than paper over it. The aggregator does not host or resell any model API; we publish neutral editorial comparisons across at least three providers and refresh prices whenever a catalog changes. The full evidence run cost well under one cent.

FAQ

How much does the MiniMax API cost? MiniMax-M2, the default text tier, costs $0.255 per million input tokens and $1.00 per million output measured via OpenRouter, versus $0.30/$1.20 on MiniMax's own platform. M1 runs higher at $0.40/$2.20 routed, and M3 lands at $0.30/$1.20 via its permanent 50%-off rate.

Why is the OpenRouter price lower than MiniMax's official price? On M2 the routed catalog rate of $0.255/$1.00 is slightly under the $0.30/$1.20 MiniMax publishes. For a US developer without a native MiniMax billing key, the route is both the available path and the cheaper one, which is why we quote routed prices throughout and flag the official figures for native re-verification.

Which is the cheapest MiniMax model? By routed input rate, minimax-01 is the floor at $0.20 input / $1.10 output, below M2's $0.255/$1.00. The trade is model age and quality, so the cheapest tier suits only workloads where the older model's output is good enough.

Is the MiniMax-M3 discount going to expire? No. The M3 promotional rate is a permanent 50%-off the raw $0.60/$2.40 card, landing at $0.30/$1.20 for inputs up to 512K. It is a standing discount, not a time-boxed sale, so it does not revert. The above-512K tier is access-limited at the full $0.60/$2.40.

How does MiniMax compare to DeepSeek on cost? DeepSeek's V4-Flash bills $0.14/$0.28 on its official endpoint, undercutting MiniMax-M2's routed $0.255/$1.00 on both sides, with the widest gap on output. DeepSeek is the cheaper raw rate; MiniMax's case rests on task fit rather than price.

Can I use MiniMax from the United States? Yes, MiniMax is reachable internationally via platform.minimax.io, and for now US developers typically reach the text models through OpenRouter rather than a native US billing key.

This is the MiniMax API pricing hub. For model-by-model rates start with MiniMax-M2 pricing, and to estimate your own spend use the MiniMax API cost calculator.

Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing and AI infrastructure benchmarking. Last verified: 2026-06-26.

MiniMax-M2, the default text tier, costs $0.255 per million input tokens and $1.00 per million output via OpenRouter, against $0.30/$1.20 on MiniMax's own…

Written by Fan Chuanyu