The Year of the Horse in 2026 has begun with MINIMAX-WP (00100) emerging as one of the most prominent highlights in the Hong Kong stock market. On February 20th, the first trading day of the lunar new year, MINIMAX-WP's share price closed up 14.52%, with its market capitalization briefly surpassing HKD 304.2 billion, setting a new record high since its listing. Since its debut on the Hong Kong Stock Exchange on January 9th at an IPO price of HKD 165 per share, the stock has achieved a maximum cumulative increase of over 480%, successfully positioning itself among the core AI-focused stocks in the Hong Kong market. This performance is not merely driven by market sentiment but is fueled by an industry-wide surge following the post-Lunar New Year release of its new-generation model, M2.5. Data from the OpenRouter platform indicates that M2.5 topped the OpenRouter popularity chart within 12 hours of its release and claimed the top spot in usage volume within a week, with weekly usage skyrocketing to 3.07 trillion tokens, surpassing the combined total of Kimi K2.5, GLM-5, and DeepSeek V3.2.

Behind this surge in usage lies a collective awakening of the Agent ecosystem. The explosive growth of M2.5 is not solely due to its pricing advantage (input cost of $0.103 per million tokens, output cost of $1.34 per million tokens, approximately 1/20th the cost of Claude Opus 4.6), but more importantly, its precise breakthrough of the "cost-capability" threshold in Agent workflows. OpenRouter's official data reveals that M2.5 leads in usage within the 100K to 1 million token long-context range, which is precisely the most representative consumption range for Agent workflows.

At the technical foundation, MINIMAX has redefined AI execution logic with its "Forge" system. Traditional models face two major challenges in Agent scenarios: context management relying on external rules leading to logical breaks, and training efficiency hampered by repetitive prefixes. MINIMAX has achieved fundamental breakthroughs through its self-developed Forge system: Architectural Decoupling: A Gateway Server isolates Agent behavior from model complexity, while a Data Pool asynchronously collects training trajectories. Training Acceleration: Prefix Tree Merging restructures training samples into a tree format, achieving approximately 40x speedup. Engineering Optimization: A Windowed FIFO strategy balances throughput efficiency and stability, and a composite reward mechanism (process reward + time reward) guides the model to select optimal execution paths.

This design directly empowers leading open-source tools. Core AgentOS frameworks such as Kilo Code and OpenClaw have fully integrated M2.5. In the SWE-Bench Verified evaluation, it achieved a pass rate of 80.2% and ranked first in multi-language tasks on Multi-SWE-Bench. The rise of AgentOS is rewriting the rules of competition for large language models. As token consumption shifts from "text interaction cost" to "action execution cost," model providers must transition from competing on "individual capabilities" to achieving "system-level compatibility." The success of MINIMAX's M2.5 validates this trend: it not only meets developer demands for a solution that "works reliably and is affordable," but also deeply integrates the model into Agent workflows through the Forge system, enabling the transition of complex task execution from demonstration to large-scale implementation.

The current strong upward trend in MINIMAX's stock price reflects market recognition of the value inherent in "AI execution layer infrastructure." Behind the 3 trillion token usage of M2.5 on OpenRouter are votes of confidence from Silicon Valley developers based on real-world needs. As the Agent ecosystem moves from concept to reality, MINIMAX is positioning itself as a technological innovator, defining the underlying logic for the next generation of AI applications. As noted by a16z research, the true competitiveness of a model lies in its "long-term fit for workloads," and MINIMAX's Forge architecture has laid crucial groundwork for this transformation.