Back to home
MiniMaxReleased 2026-06-01

MiniMax M3

MiniMax M3 is MiniMax's open-source multimodal model released June 1, 2026. 1M context, native multimodal (text+image+audio+video), MSA sparse attention architecture. $0.30-$0.60 per 1M input tokens.

Context
1M tokens
Input
$0.6/1M
Output
$2.4/1M
Modalities
textimageaudiovideo
💡 Promo: $0.30/$1.20. International: $0.60/$2.40.

Best For

Multimodal at low priceNative sparse attentionOpen-source

Strengths

  • Open-source under Apache 2.0
  • Native multimodal (text+image+audio+video)
  • MSA sparse attention = efficient inference
  • 1M context window
  • Cheapest multimodal with video understanding

Weaknesses

  • Less ecosystem support than GPT-5.6
  • English slightly behind Western models
  • Some advanced features require Pro tier

Related Alternatives

Gemini 3.6 Flash

Multimodal, more mature

Qwen3.8-Max

Larger, Chinese-optimized

Frequently Asked Questions

Is MiniMax M3 open source?

Yes, MiniMax M3 was open-sourced June 1, 2026 under Apache 2.0. You can self-host, fine-tune, and use commercially. Weights available on Hugging Face.

How much does MiniMax M3 cost via API?

MiniMax M3 API costs $0.60 per 1M input and $2.40 per 1M output internationally. A promotional rate of $0.30/$1.20 is available. In China: ¥1/¥4 per 1M tokens.

See MiniMax M3's full record in our database

Live pricing for all MiniMax plans, token cost calculator, and the latest price changes.

View Full Comparison →