minimax
MiniMax is a leading multimodal AI company in China that develops the M series of large language models. The M3 flagship supports 1M extra-long context, while the M2.5 series covers standard and Lightning (low-latency) variants. It is known for its excellent value, strong Chinese-language understanding, and multimodal capabilities (speech, video).
1M token context window (M3); visual understanding; speech synthesis and recognition (MiniMax speech engine); video generation; Function Calling; JSON mode; context caching (cache_read 20% of the input price, with M2.5 supporting cache_write); OpenAI-compatible interface.
High-value reasoning and conversation; long-document analysis; speech synthesis and cloning; video content understanding; agent workflows; customer support and chatbots; creative writing.
The Lightning variant doubles the output price; some models are available only in China; speech/video are billed per use; no on-premises deployment; the enterprise edition requires contacting sales.
| Model | Input / 1M | Output / 1M | Cache Read / 1M | Per 1M |
|---|---|---|---|---|
| 1M | ||||
| 1M | ||||
| 1M | ||||
| 1M |