Event date · · MiniMax
MiniMax-M1 Open Source: Domestic Reasoning Model Competition Extends to Million-Token Context
MiniMax released and open-sourced the M1 hybrid attention reasoning model, providing million-token input context.
Domestic model vendors are beginning to differentiate along the axes of reasoning, long context, Agent tool use, and cost.
Lightning Attention and hybrid attention reduce long-context computational cost, targeting complex productivity tasks.
Domestic frontier models are shifting from chasing single benchmarks to jointly optimizing architecture and product constraints.
Long documents, codebases, and enterprise knowledge scenarios gain a lower-cost domestic model option.
Observe open-source reproduction, long-context effectiveness, API usage, and Agent product conversion.