MiniMax

Editorial Review

MiniMax has quietly become one of the most technically interesting multimodal labs operating outside the US/EU. Hailuo AI gained global traction in 2024 as a Sora alternative — the consumer web app produces strikingly cinematic video clips from text or image prompts, and the free daily quota made it a viral favorite for creators. The Standard plan at roughly $19.99/mo unlocks higher-resolution outputs and priority generation. On the language model side, MiniMax-M1 is the standout: a 456B-parameter Mixture-of-Experts model with 1M token context, released with open weights under a permissive license. It benchmarks competitively with Western reasoning models and offers a rare combination of frontier-scale architecture with self-hostability — appealing for organizations that want long-context inference but can't send data to closed API providers. Considerations are similar to other China-based labs: data practices fall under Chinese jurisdiction for hosted services, though self-hosting M1 sidesteps this entirely. The English-language documentation and developer tooling are improving but remain less polished than OpenAI or Anthropic. For video specifically, Hailuo is genuinely competitive with Runway Gen-3, Kling, and Pika — and the free tier makes it the easiest way to evaluate AI video without a credit card.

Key Features