All models

MiniMax M3: a 1M-token vision and code model

Official MiniMax M3 routing for coding agents, 1M context, and native multimodal understanding.

1M tokens · Text / Vision / Code · Prompt cache

Okou no longer runs MiniMax M3. This page is kept as a reference for its specs, pricing and benchmarks. For the same kind of work, use GPT 5.6 Luna.

See GPT 5.6 Luna

What is MiniMax M3?

June 1, 2026 · Official MiniMax M3 text model alongside the retained M2 series.

MiniMax M3 is the new official MiniMax text model for coding and agentic workloads. MiniMax describes it as combining frontier coding capability, a 1M-token context window, and native multimodal understanding.

The existing M2 entries stay available, while M3 is the MiniMax provider default for new selections.

What's notable about MiniMax M3

Headline architecture and capability features.

M3 uses MiniMax Sparse Attention for ultra-long context. The official API supports up to 1M tokens with a 512K guaranteed minimum, automatic prompt cache support, and native multimodal understanding.

Specs at a glance

FamilyMiniMax M3
ModalitiesText, vision, code
LanguagesMultilingual
Context window1M tokens (512K guaranteed minimum)
Prompt cachingSupported with automatic cache reads

MiniMax M3 benchmarks

PostTrainBench Livevendor reported
Rank #3
BrowseCompvendor reported
83.5

MiniMax M3 pricing

Provider list price, per 1M tokens.

Input$0.60
Output$2.40
Cache read$0.12
Cache writeNot billed

How MiniMax M3 behaves in practice

Observed behaviour from production agent runs.

Coding agents

The best MiniMax option for coding assistants, long tool chains, and multi-step repository work.

Long context

A 1M context window lets agents keep large documents, code, logs, and previous work in scope without switching model families.

Multimodal

Native vision understanding makes M3 a better fit than M2 when screenshots, diagrams, or visual artifacts are part of the workflow.

Best agent tasks for MiniMax M3

The coding agent that must stay on MiniMax

Repository edits, debugging, and agentic coding flows where your deployment already uses MiniMax keys but needs a stronger coding model than M2.1.

The long-context review run

Large pull requests, logs, notebooks, or design documents that should stay in one model context while the agent reasons and edits.

MiniMax M3 vs other models

MiniMax M3 vs Claude Sonnet 4.6

Sonnet 4.6 remains the baseline for reliability on complex English tool-use. M3 is much cheaper and attractive for MiniMax-native coding agents, but should be validated on critical workflows.

Bottom line: should you use MiniMax M3?

Use MiniMax M3 when you want the official MiniMax coding model with long context and vision support. Keep M2.1 when cost and compatibility matter more.

Availability of MiniMax M3 on Okou

MiniMax M3 has been removed from the Okou lineup, so it can no longer be selected in chat or in a workflow, and it is not available with your own API key. GPT 5.6 Luna covers the same cost-saving tier.