Skip to content
#12A+Jun 1, 2026Evidence B

MiniMax M3

MiniMax · 428B total / 23B active (MoE)

Open weights with 1M context and strong agents

A 428B-A23B multimodal model released 1 June 2026 with open weights and 1M context. Agents 93, ecosystem 92 and perception 94 give a composite of 91.1; Chinese 88 is its clear gap versus the domestic top tier. The community licence permits commercial use but is not OSI open source.

TextImageVideoNative

Last updated Oct 21, 2026Editorial scoring

Bottom line

MiniMax M3 suits “long context plus tool use”: 1M context, agents 93 and ecosystem 92 are rare among open weights. Be careful with Chinese-language work — 88 means localisation and cultural context still need post-processing — and clear the community licence with legal first.

Eight-dimension breakdown

editorial mapping · 0–100
Perception94Document92Chart QA90Video89Agent93Generation84Chinese88Ecosystem928D
  • MiniMax M391.1
Visual perception & grounding20%
94
Document parsing & Chinese OCR18%
92
Chart & document reasoning14%
90
Video & long-horizon reasoning14%
89
GUI agents & tool use14%
93
Generation & editing6%
84
Chinese context & culture6%
88
Ecosystem, cost & availability8%
92

Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.

Public benchmarks

as reported, not converted
Benchmark / metricValueNote
参数规模428B-A23BOfficial spec
上下文1MOfficial spec
许可证社区许可Commercial use, not OSI

Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.

Scenario mapping

five stress tests, one scale
ScenarioScoreObservation
Chinese receipt OCR90 · Fine on clean receipts
Chart data extraction89 · Stable on standard charts
30-min video QA88 · Long video works; event localization is average
Mobile GUI control92 · Good tool calls and recovery
Chinese cultural context86 · General phrasing fine; weak on obscure allusions

Strengths and limitations

Strengths

  • 1M context plus agents 93 puts it near the front of open weights for long tasks and tool orchestration.
  • 428B-A23B capacity with perception 94 and documents 92 keeps general tasks stable.
  • Ecosystem 92: docs, SDKs and community reproduction are relatively complete.

Limitations

  • Chinese 88: idioms, memes and official register clearly trail Qwen and Doubao.
  • The MiniMax community licence is not OSI open source and carries extra commercial terms.
  • Video 89 and generation 84: long video and creative generation are not strengths.

Price and real cost

List price

¥4/M tokens

≈ ¥4 / 1M output tokens; weights free. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥2.4 (assuming input at 25% of output price).

Use your volume

About MiniMax M3

Is MiniMax M3 open source?

Licence: Open weights (MiniMax community licence). Weights are available, but check the exact terms and filing requirements before commercial use.

How much does MiniMax M3 cost?

≈ ¥4 / 1M output tokens; weights free. It ranks #12 with a composite of 91.1. Real bills depend on resolution, tiling, caching and volume discounts.

What is MiniMax M3 best for?

Long-context agent workflows; Tool-heavy automation; Self-hosted long-document analysis。

Why does MiniMax M3 score 91.1?

The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Visual perception & grounding; its weakest is Generation & editing.

Personal ranking · Not an official score

Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.