MINIMAX-WP (00100) saw its shares rise more than 10%. As of the time of writing, the stock was up 10.37%, trading at HKD 1,048, with a turnover of HKD 580 million.
Recently, MiniMax unveiled its new flagship Agent model, M2.7, which for the first time demonstrates a "self-evolving model" pathway. By establishing an Agent Harness system, the model deeply engages in its own training and optimization processes, handling 30% to 50% of workloads in certain R&D scenarios and achieving approximately 30% improvement on internal evaluation benchmarks.
Notably, the domestic large model MiniMax2.7 has recently become a focal point in the tech sector, with its open-source plan officially coming to light. According to disclosures from the Huggingface community, the model will have its source code released to global developers this weekend. This move is seen as a significant breakthrough in reducing the costs of AI application development. Previously, the model garnered widespread attention for its deep integration with the OpenClaw plugin, with the developer community widely anticipating more flexible application scenarios post-open-sourcing.
An internal source revealed that the open-source version will reduce inference costs by 70% through a distributed computing architecture. In the MMClaw evaluation, the performance gap between this model and the latest Sonnet4.6 has narrowed to just 8%, yet its computational cost per unit is only one-third of the latter. This cost-performance advantage gives it significant competitiveness in the small and medium-sized enterprise market.
Comments