Another "big bomb" has emerged in the domestic AI field. On February 12, 2026, MiniMax officially launched its latest flagship programming large model MiniMax M2.5. Officially defined as "the world's first production-level model designed natively for Agent scenarios," the release not only caused a stir in the tech community but also directly ignited the capital market.

Performance Peak: Directly Competing with Claude Opus 4.6

MiniMax M2.5 demonstrates strong competitiveness in programming and agent performance (Coding & Agentic):

  • Full-Stack Programming: Supports full-cycle development for PC, App, and cross-platform applications, showcasing profound programming expertise.

  • Productivity King: In core office scenarios such as advanced Excel processing, in-depth research, and PPT creation, it is at the industry-leading (SOTA) level.

  • International Standards: The official clearly states that the comprehensive performance of this model directly competes with top international models like Claude Opus 4.6.

Efficiency Myth: Small Parameters, High Throughput

Differing from blindly pursuing parameter scale, M2.5 has taken a highly cost-effective approach:

  • Streamlined Architecture: It has only 10B activation parameters, significantly reducing memory usage.

  • High Energy Efficiency: It supports an impressive 100 TPS throughput, with inference speed even surpassing some top international models.

Capital Mania: Market Value Surpasses 180 Billion HKD

Stimulated by the major positive news of the new model's release, MiniMax saw its share price rise by more than 20% during the day, with its total market value successfully exceeding 180 billion HKD. This impressive performance once again solidified its strong position in the competition for the title of "Top AI Stock" against competitors like Zhipu.

With the launch of MiniMax M2.5, domestic large models are transitioning from "general追赶" to "professional scenario reshaping," marking the beginning of a new era for AI programming and Agent applications.