Model release · MiniMaxNew

MiniMax M3.1 Flash Pricing vs M3 & DeepSeek V4.1 Flash

Prices verified Oct 21, 2026/Price version 2026-10-21.1/Released Oct 19, 2026/Sources

MiniMax-M3.1-Flash is MiniMax’s Oct 19, 2026 Flash model: a 1M-token context, text, image and video input, and thinking that’s always on. It costs $0.15 per 1M input tokens ($0.015 on a cache hit) and $0.60 per 1M output tokens at any prompt length: half MiniMax M3’s input and output price, and the same as DeepSeek V4.1 Flash’s off-peak input and output price.

Calculate your cost Compare more models

On this page

01API pricing
02Cost calculator
03Cost per task
04Use the API
05FAQ
—Sources & change log

01 — API pricing

MiniMax M3.1 Flash API pricing vs M3 and DeepSeek V4.1 Flash

MiniMax API · MiniMax-M3.1-Flash · USD per 1M tokensVerified Oct 21, 2026

API prices per 1M tokens for MiniMax M3.1 Flash, MiniMax M3 and DeepSeek V4.1 Flash
Tokens	M3.1 FlashAny length	M3≤ 512,000 in	M3> 512,000 in	DeepSeek V4.1 FlashPeak	DeepSeek V4.1 FlashOff-peak
Inputnew tokens	$0.15	$0.30	$0.60	$0.30	$0.15
Inputcache hit	$0.015	$0.06	$0.12	$0.006	$0.003
Outputincludes thinking tokens	$0.60	$1.20	$2.40	$1.20	$0.60

M3.1 Flash costs half M3’s input and output price and a quarter of its cache-read price.
Past 512,000 input tokens, M3 bills the whole request at $0.60, $0.12 and $2.40 per 1M. M3.1 Flash has one price at any length.
Input and output cost the same as DeepSeek V4.1 Flash’s off-peak price and half its peak price.
DeepSeek’s cache hits cost less: $0.006 at peak and $0.003 off-peak, against $0.015.

Model ID: MiniMax-M3.1-Flash
Context window: 1,000,000 tokens
Max outputhighest max_completion_tokens: 524,288 tokens
Cache writes: No charge · hits and new input only
Long-context premium: None
Thinkingcan’t be turned off: Always on · default effort max
Effort levels: low · medium · high · xhigh · max
Recommended max outputmax_completion_tokens: 131,072 tokens
Prompt caching: Automatic · prompts of 512+ tokens
Rate limitper account: 200 RPM · 10,000,000 TPM
Input: Text, images and video
Open weights: Not yet · MiniMax says later
Released: Oct 19, 2026

MiniMax M3 also has a Priority tier (service_tier: "priority") for faster, more reliable admission, at 1.5× these rates. We price the standard tier.

02 — Cost calculator

Cost calculator

Preset to MiniMax M3.1 Flash. Same calculator as the LLM API cost calculator.

Model

Only models with verified prices are listed. A model is added after we check the vendor’s price page.

Tokens per call

SimpleDetailed

MiniMax doesn’t bill cache writes separately, so both modes treat them as uncached input.

Total input tokens

tokens

Everything you send: system prompt, history, tools and user text.

Cache hit rate

% of input

Share of input tokens read from cache — not share of requests.

Output tokens

tokens

Includes reasoning tokens — don’t add them twice.

= 2,000 uncached + 0 cache-read tokens per call.

Volume

Calls per day

API calls

API calls — not users or conversations.

Days in period

days

The period your total covers.

Estimate

MiniMax M3.1 Flash

MiniMax API · Standard · MiniMax-M3.1-Flash

1,000 API calls/day × 30 days

Uncached input$9.002,000 tok × $0.15 / 1M = $0.0003000 per call

Cache read—0 tokens

Cache writesNo chargeMiniMax caches prompts automatically and bills new input, cache hits and output only; there’s no cache-write charge.

Output · incl. reasoning$36.002,000 tok × $0.60 / 1M = $0.001200 per call

Per call$0.001500

Period total30 days · 30,000 API calls$45.00

Monthly budget$45.00

Uncached input20.0%
Cache read0.0%
Cache write0.0%
Output80.0%

Excludes tax. Totals are summed before rounding.

Official price sources

Prices verified 2026-10-21Price version 2026-10-21.1USD

Output includes thinking tokens. Thinking is always on and defaults to effort max; MiniMax counts its tokens in completion_tokens and bills them as output.
Copying from MiniMax’s usage in Detailed mode? OpenAI format: prompt_tokens_details.cached_tokens is Cache read, prompt_tokens minus cached_tokens is Uncached input and completion_tokens is Output. Anthropic format: cache_read_input_tokens is Cache read and input_tokens is Uncached input.

Same usage on M3 and DeepSeek V4.1 Flash

Uses the tokens and volume above. DeepSeek V4.1 Flash bills less off-peak, so it gets a row at each price. Moving a warm cache over from M3 or DeepSeek? Choose Same cache split.

Cache on other models

Don’t carry overSame cache split

Model	Applied price / 1M · in · out	Per call	Period · 30 days	vs MiniMax M3.1 Flash
MiniMax M3.1 FlashSelected · MiniMax	$0.15 · $0.60	$0.001500	$45.00	Selected
DeepSeek V4.1 FlashDeepSeek · Peak	$0.30 · $1.20	$0.003000	$90.00	+$45.00 · +100.0%
DeepSeek V4.1 FlashDeepSeek · Off-peak	$0.15 · $0.60	$0.001500	$45.00	±$0.00 · 0.0%
MiniMax M3MiniMax	$0.30 · $1.20	$0.003000	$90.00	+$45.00 · +100.0%

Default: cache reads are priced as new input on other models — a cache hit rate doesn’t carry over to a model you haven’t run.
Each model’s tokenizer is different, so the same text has a different token count. Equal token counts here show a price scenario only.
Displayed input and output rates include any long-context premium for this request size.
Tool calls such as web search aren’t included in any row.

Compare more models

03 — Cost per task

Cost per task: M3.1 Flash vs DeepSeek V4.1 Flash 1017

MiniMax-reported · release charts, Oct 19, 2026MiniMax’s charts ↗

Score and approximate cost per task at effort max, MiniMax M3.1 Flash against DeepSeek V4.1 Flash 1017, from MiniMax’s charts
Benchmark	M3.1 FlashScore	M3.1 FlashPer task	DeepSeek 1017Score	DeepSeek 1017Per task
Automation-Bench v1.0.6Claude Code harness	62.8%	about $0.045	60.5%	about $0.09
DeepSWE v1.1mini-swe-agent harness	73.7%	about $0.28	77.0%	about $0.20

MiniMax published these charts with the release. Scores are as labelled or published; we read the costs off log-scale axes, so treat them as approximate. They’re never used in a calculation.
MiniMax priced DeepSeek at peak rates on its Automation-Bench v1.0.6 chart. Off-peak halves DeepSeek’s cost per task, to about $0.045 — level with M3.1 Flash.
At effort low, the same chart puts M3.1 Flash at 52.8% for about $0.027 a task. Leaving effort out means max.
MiniMax M3 isn’t on these charts. MiniMax says M3.1 Flash cuts usage costs on agentic tasks by 70% against its predecessor, without a chart for it.

04 — Use the API

How to use MiniMax M3.1 Flash through the API

Anthropic-compatiblerecommended by MiniMax: https://api.minimax.io/anthropic
OpenAI-compatiblebase_url: https://api.minimax.io/v1
Model name: MiniMax-M3.1-Flash
Also accepts: Responses API

Quick start

curlPythonNode.js

curl https://api.minimax.io/anthropic/v1/messages \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $MINIMAX_API_KEY" \
  -d '{
    "model": "MiniMax-M3.1-Flash",
    "max_tokens": 4096,
    "output_config": {"effort": "high"},
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Thinking depth by request format

The effort field, where thinking comes back and the output limit, by request format
Request format	Effort	Thinking comes back as	Output limit
Anthropic-compatible	`"output_config": {"effort": "high"}`	thinking content block	`max_tokens`
OpenAI-compatible	`"reasoning_effort": "high"`	reasoning_content field	`max_completion_tokens`
OpenAI Responses	`"reasoning": {"effort": "high"}`	reasoning output item	`max_output_tokens`

Effort takes low, medium, high, xhigh, max. Leave it out and you get max, which thinks longest and bills the most output tokens. Thinking can’t be turned off: "effort": "none" or a disabled thinking setting returns a 400 error ("requires adaptive thinking"), so lower the effort instead. MiniMax’s Model Invocation guide ↗

On OpenRouter the model is minimax/minimax-m3.1-flash, billed at MiniMax’s list price. Its free variant, minimax/minimax-m3.1-flash:free, ran only for the first 72 hours, until Oct 22, 2026, 12:00 UTC.

Coding tools · MiniMax’s setup guides

Claude Code ↗Cursor ↗Codex ↗Other tools ↗

Rather use an app? MiniMax Code, MiniMax’s desktop coding agent, runs M3.1 Flash too: Download MiniMax Code ↗

05 — FAQ

MiniMax M3.1 Flash FAQ

What is MiniMax M3.1 Flash?

MiniMax M3.1 Flash (API name MiniMax-M3.1-Flash) is the Flash model MiniMax released on Oct 19, 2026, aimed at agentic tasks, coding, computer use and cybersecurity. It takes text, images and video, has a 1M-token context and always thinks before it answers. It’s on the MiniMax API, the M Plan subscription and MiniMax Code.

How much does the MiniMax M3.1 Flash API cost?

$0.15 per 1M input tokens, $0.015 per 1M cached input tokens and $0.60 per 1M output tokens, at any prompt length. Thinking tokens bill as output, and MiniMax doesn’t charge for writing to its cache.

How much cheaper is M3.1 Flash than MiniMax M3?

M3.1 Flash costs half M3’s input and output price and a quarter of its cache-read price: $0.15, $0.015 and $0.60 per 1M tokens against M3’s $0.30, $0.06 and $1.20. Past 512,000 input tokens M3 bills the whole request at $0.60, $0.12 and $2.40, while M3.1 Flash keeps one price at any length: a quarter of M3’s price there.

Is MiniMax M3.1 Flash cheaper than DeepSeek V4.1 Flash?

It depends on the hour and your cache hits. M3.1 Flash’s input and output cost the same as DeepSeek V4.1 Flash’s off-peak rates and half its peak rates, but DeepSeek’s cache hits cost less: $0.006 at peak and $0.003 off-peak, against $0.015. With a mostly cached prompt, DeepSeek off-peak can come out cheaper. Try your own usage in the calculator above, which shows DeepSeek at both prices.

Can I turn off thinking in MiniMax M3.1 Flash?

No. Thinking is always on, and asking for "effort": "none" or turning thinking off returns a 400 error. To spend fewer tokens, lower the effort: low, medium, high, xhigh, max. Leave it out and you get max.

Is MiniMax M3.1 Flash the Space Bunny model?

Yes. MiniMax says Space Bunny, the anonymous model that processed more than 200 trillion tokens before launch, was an early version of M3.1 Flash.

Is MiniMax M3.1 Flash free or open source?

Neither, for now. The MiniMax API bills it at the prices above, and OpenRouter’s free variant ran only for the first 72 hours, until Oct 22, 2026, 12:00 UTC. MiniMax calls M3.1 Flash an open-weight model but says the weights will be released later.

Sources

Sources & change log

Official pages · checked Oct 21, 2026

MiniMaxMiniMax M3.1 Flash release, Oct 19, 2026
MiniMaxPricing Overview
MiniMaxModel Invocation
MiniMaxCreate Chat Completion
MiniMaxPrompt Caching
MiniMaxRate Limits

Elsewhere · OpenRouter’s free launch window

OpenRouterMiniMax M3.1 Flash (free) launch window

DeepSeek V4.1 Flash prices and sources are on our DeepSeek V4.1 Flash page.

Change log

Oct 21, 2026Page published for MiniMax M3.1 Flash. Cost per task from MiniMax’s release charts; prices checked against MiniMax’s Pricing Overview, price version 2026-10-21.1.

Pilot

Already paying an LLM API bill?

Join the pilot for project cost reviews and budgets. Two questions first — we only ask for your email if it fits.