mlx-optiq
← Models

MiniMax M3.1 Flash

MiniMax M3.1 Flash on OptiQ Cloud, billed per token.

cloud

MiniMax M3.1 Flash is a multimodal model from MiniMax: it reads text, images and video, with a context window of about 1 million tokens. It is served by OptiQ Cloud and billed per token. All OptiQ Cloud models are served at 8-bit precision or better (int8, fp8, fp16, bf16 or fp32), and your requests are never used for training.

Input$0.15 / M
Cached input$0.02 / M
Output$0.60 / M
Context1M tokens

Next to the other OptiQ Cloud models

Scores are from each maker's model card, for benchmarks that more than one card reports; OptiQ did not run them. Prices are what OptiQ Cloud charges.

DeepSWE v1.1

DeepSWE v1.1, as each maker published itGLM 5.3 Flash 63.4; DeepSeek V4.1 Flash 77.0; GLM 5.3 66.9; MiniMax M3.1 Flash 73.7GLM 5.3 Flash63.4DeepSeek V4.1 Flash77.0GLM 5.366.9MiniMax M3.1 Flash73.7

Output price, $ per million tokens

Output price, $ per million tokensGLM 5.3 Flash $0.50; DeepSeek V4.1 Flash $0.50; GLM 5.3 $4.40; MiniMax M3.1 Flash $0.60GLM 5.3 Flash$0.50DeepSeek V4.1 Flash$0.50GLM 5.3$4.40MiniMax M3.1 Flash$0.60

Input price, $ per million tokens

Input price, $ per million tokensGLM 5.3 Flash $0.15; DeepSeek V4.1 Flash $0.10; GLM 5.3 $1.40; MiniMax M3.1 Flash $0.15GLM 5.3 Flash$0.15DeepSeek V4.1 Flash$0.10GLM 5.3$1.40MiniMax M3.1 Flash$0.15

How MiniMax compares it

Scores published by MiniMax for MiniMax M3.1 Flash and other models, taken from the comparison table on MiniMax's model card. Higher is better; OptiQ did not run these benchmarks.

DeepSWE v1.1

DeepSWE v1.1, as published by MiniMaxMiniMax M3.1 Flash 73.7; DS-V4.1-Flash-1017 77.0; DS-V4.1-Flash 74.2; Kimi K3 67.5; GLM-5.3 66.9; GPT-5.6 Sol 73.0; Opus 5 74.0MiniMax M3.1 Flash73.7DS-V4.1-Flash-101777.0DS-V4.1-Flash74.2Kimi K367.5GLM-5.366.9GPT-5.6 Sol73.0Opus 574.0

Terminal Bench 4.0 (Codex)

Terminal Bench 4.0 (Codex), as published by MiniMaxMiniMax M3.1 Flash 38.1; DS-V4.1-Flash-1017 42.3; DS-V4.1-Flash 31.2; Kimi K3 12.6; GLM-5.3 37.9; GPT-5.6 Sol 39.9; Opus 5 51.8MiniMax M3.1 Flash38.1DS-V4.1-Flash-101742.3DS-V4.1-Flash31.2Kimi K312.6GLM-5.337.9GPT-5.6 Sol39.9Opus 551.8

OSWorld 2.0

OSWorld 2.0, as published by MiniMaxMiniMax M3.1 Flash 61.7; DS-V4.1-Flash 54.3; Kimi K3 58.3; GPT-5.6 Sol 65.7; Opus 5 70.6MiniMax M3.1 Flash61.7DS-V4.1-Flash54.3Kimi K358.3GPT-5.6 Sol65.7Opus 570.6

LOCA-Bench 256K

LOCA-Bench 256K, as published by MiniMaxMiniMax M3.1 Flash 60.3; Kimi K3 49.7; GLM-5.3 50.3; GPT-5.6 Sol 60.0; Opus 5 53.7MiniMax M3.1 Flash60.3Kimi K349.7GLM-5.350.3GPT-5.6 Sol60.0Opus 553.7

Source: MiniMaxAI/MiniMax-M3.1-Flash

Use it

Connect this Mac

Sign in once. New accounts get $2 of credit. See OptiQ Cloud.

pip install mlx-optiq
optiq cloud login

Claude Code

optiq run claude --cloud minimax-m3.1-flash

Codex

optiq run codex --cloud minimax-m3.1-flash

OpenCode

optiq run opencode --cloud minimax-m3.1-flash

Mistral Vibe

optiq run vibe --cloud minimax-m3.1-flash

Hermes Agent

optiq run hermes --cloud minimax-m3.1-flash

OpenClaw

optiq run openclaw --cloud minimax-m3.1-flash

No local model is loaded. optiq run starts optiq serve --cloud-only, sets the agent up for this session only, and stops the server when the agent exits.

OptiQ Code

Boost: this model leads the task and your local model is its helper.

/boost minimax-m3.1-flash

OptiQ Lab, and any app that talks to optiq serve

Pick optiq-cloud/minimax-m3.1-flash in the model list. With no local model:

optiq serve --cloud-only

The API, from anywhere

Create a key on the dashboard.

curl https://cloud.mlx-optiq.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OPTIQ_CLOUD_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "minimax-m3.1-flash",
       "messages": [{"role": "user", "content": "Hello"}]}'
from openai import OpenAI

client = OpenAI(
    base_url="https://cloud.mlx-optiq.com/api/v1",
    api_key="<your key>")
reply = client.chat.completions.create(
    model="minimax-m3.1-flash",
    messages=[{"role": "user", "content": "Hello"}])
print(reply.choices[0].message.content)

Published benchmarks

These coding and agent benchmark figures were published by MiniMax on its official model card (MiniMaxAI/MiniMax-M3.1-Flash).

DeepSWE v1.173.7
Terminal Bench 4.0 (Codex)38.1
OSWorld 2.061.7
LOCA-Bench 256K60.3

Details

Input$0.15 per million tokens
Cached input$0.02 per million tokens
Output, including reasoning$0.60 per million tokens
Context window1M tokens
Longest reply524,288 tokens
Model idminimax-m3.1-flash on the API, optiq-cloud/minimax-m3.1-flash in optiq serve