MiniMax

Overview

MiniMax is a Shanghai-based AI foundation model company founded in 2021 by Yan Junjie and Zhou Yucong, both former SenseTime computer vision experts. The company name derives from the min-max algorithm in game theory, which searches for optimal solutions in worst-case scenarios. MiniMax went public on the Hong Kong Stock Exchange in January 2026 as one of China's "AI Tiger" startups.

The company develops proprietary multimodal models and consumer-facing AI products including MiniMax Agent, Hailuo AI (video generation), MiniMax Audio, and Talkie (AI companion app).

Text Models

ModelDescription
MiniMax-M2.1230B total params (10B activated), optimized for code generation and refactoring
MiniMax-M2.1-lightningSame performance as M2.1 with faster inference
MiniMax-M2200k context, 128k output, agentic capabilities, function calling (open weights)
MiniMax-M2-herRole-playing and multi-turn conversation specialist
MiniMax-M1456B params, reasoning model with Lightning Attention (40k/80k context, open weights)
MiniMax-Text-01456B params, 4M token context window (open weights)
MiniMax-VL-01456B params, vision-language multimodal model (open weights)

Speech Models

ModelDescription
Speech 2.8 HD/TurboLatest speech model with tonal nuance and timbre similarity, 40 languages
Speech 2.6 HD/TurboHigh-quality text-to-speech with voice cloning, 40 languages, 7 emotions
Speech 02 HD/TurboEarlier generation with 24 languages, 7 emotions

All speech models support voice cloning, emotional expression, and multiple dialects.

Video Models (Hailuo)

ModelDescription
Hailuo 2.3Text-to-video and image-to-video, 1080p 6s or 768p 6-10s at 24fps
Hailuo 2.3 FastImage-to-video with faster generation
Hailuo 02Earlier version with 512p-1080p support

Music Models

ModelDescription
Music 2.5Text-to-music with studio-quality output and expressive vocals
Music 2.0Enhanced musicality with natural vocals

Consumer Products

  • MiniMax Agent: General-purpose AI assistant
  • Hailuo AI: Video generation platform (hailuoai.video)
  • MiniMax Audio: Text-to-speech and voice cloning
  • Talkie: AI companion app for character conversations

Open Weights

MiniMax releases several models as open weights on Hugging Face:

  • MiniMax-M2 (229B parameters)
  • MiniMax-M1 (456B parameters, 40k and 80k context variants)
  • MiniMax-Text-01 (456B parameters)
  • MiniMax-VL-01 (456B parameters)
  • SynLogic reasoning models (7B and 32B)

Links

Models in this family

Feb 12, 2026
Jan 23, 2026
Dec 23, 2025
Oct 23, 2025