M3.1 Flash
Try it

MiniMax
M3.1 Flash

MiniMax’s new fast model. It builds a full web page for under 2¢.

See the buildsTry it on Together AI ↗
Aurora lodge, built by MiniMax M3.1 Flash
Aurora lodge · $0.010 · 78.3sOpen ↗
$0.15 / $0.60
per 1M tokens, in / out
1M
tokens of context
#5
of 174 on Artificial Analysis
330 tok/s
on Together AI

117× cheaper than Opus 5.5.

Same prompt, two models, 11 times. Open any build and click around.

vs Opus 5.5

117× cheaper

5 builds, faster on all 5

vs GPT-6 Astra

76× cheaper

3 builds, faster on all 3

vs DeepSeek V4.1 Flash

2.5× cheaper

8 builds, slower on 7

Aurora lodge

Prompt: A landing page for a lodge in Finnish Lapland where every room is a glass igloo for watching the northern lights. The hero should be a night sky with a living aurora rippling over snowy pines. Below it: the rooms, what a night there feels like hour by hour, a small aurora forecast for the next few nights, and a booking section where picking dates updates the price. Make it feel cold, quiet and expensive.

MiniMax M3.1 Flash$0.010 · 78.3sOpen ↗
aurora-lodge built by MiniMax M3.1 Flash
Claude Opus 5.5$1.62 · 723sOpen ↗
aurora-lodge built by Claude Opus 5.5
M3.1 Flash Opus 5.5 GPT-6 Astra DeepSeek V4.1 Flashcost per build, log scale
$0.01$0.1$1

Cost is list price for the tokens each run used. Times come from MiniMax’s own API at launch, which ran at about 154 tok/s; Together AI serves it at about 330.

22¢ per task.

Cheaper than every other fast model here, and #5 of 174 on intelligence.

Cost to finish one Intelligence Index task
  • 1MiniMax M3.1 Flash$0.22
  • 2DeepSeek V4.1 Flash$0.26
  • 3GLM-5.3 Flash$0.28
  • 4MiniMax M3$0.51
  • 5Gemini 3.8 Flash$1.24

Artificial Analysis Intelligence Index and cost to run it, October 2026.

What a month costs.

Pick a workload or set your own. Input it has seen before is 90% off.

Your monthly bill
Requests a day2,000
Tokens in, per request40,000
Tokens out, per request2,500
Input served from cache85%

On M3.1 Flash

360M fresh in × $0.15
$54.00
2,040M cached in × $0.015
$30.60
150M out × $0.60
$90.00
Total$175

Same traffic elsewhere, no caching

  • MiniMax M3.1 Flash$450
  • DeepSeek V4.1 Flash$900
  • Kimi K3$9,450
  • Claude Opus 5.5$12,600
  • GPT-6 Astra$31,500

28× less than Opus 5.5, before caching saves another $275.

Run it on Together AI.

Python, TypeScript or cURL. Each one runs as is.

  • pip install together
  • export TOGETHER_API_KEY=your_key
from together import Together

client = Together()

resp = client.chat.completions.create(
    model="MiniMaxAI/MiniMax-M3.1-Flash",
    reasoning_effort="high",
    max_tokens=32000,  # thinking counts toward this
    messages=[{"role": "user", "content": "Build a landing page for a ski lodge"}],
)
print(resp.choices[0].message.content)

What it's good at.

  • Bug fixes and features

    73.7%

    on DeepSWE. Opus 5 scores 74%.

  • Agents and tool calls

    76.2

    on BankerToolBench, level with Opus 5.

  • Landing pages and UI

    1.4¢

    the average cost of our 11 arena builds.

  • Long docs, images, video

    1M

    tokens of context at one flat price.

Try M3.1 Flash.

MiniMaxAI/MiniMax-M3.1-Flash

Get an API keyOpen the playground