Overview
MiniMax-M1 combines mixture-of-experts with linear “lightning attention”, which keeps compute low for very long inputs and long chains of thought.
Highlights
- 456B total, 45.9B active parameters
- 1M-token context window
- Thinking budgets of 40K and 80K tokens
- Apache 2.0 license
Typical uses
Long-document reasoning, software engineering agents.
Things to consider
- Data-center hardware required to self-host.
Specifications are taken from the model card on Hugging Face and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.