MiniMax Prepares to Launch Next-Generation M3 Large Language Model

MiniMax says its next-generation M3 large language model is in the final preparation stage for launch.

M3 uses a custom sparse attention design with an Index Branch and Sparse Branch to improve long-context processing.

The company claims major speed gains over M2 and lower enterprise costs for million-token workloads.

MiniMax has not yet revealed the release date, technical specs, or parameter count.