- MiniMax will implement dynamic traffic limits during peak hours to manage an unexpected surge in demand for its M2.7 model.
- The Chinese AI startup's new self-evolving model claims to rival top competitors in key coding benchmarks.
MiniMax will implement dynamic traffic limits during peak hours after the popularity of its newly released M2.7 model exceeded expectations.
The Chinese artificial intelligence company announced the adjustment on its open platform on Friday, a move aimed at ensuring service stability and availability for all users.
The firm observed that some requests stemmed from highly concurrent automated batch tasks, as well as multi-user sharing modes, according to the announcement.
To prevent abnormal traffic from crowding out the public computing pool, MiniMax will adjust rates based on account usage to ensure a stable experience for the majority of users.
Peak hours typically occur between 15:00 and 17:30 on weekdays, the company said. Starter and Plus users will be limited to about one continuous Agent call during this period.
Max users will be allowed about two Agents, while Ultra users will get about four. The company is also setting weekly usage quotas for new users.
Customers who purchase after March 23, 2026, will be subject to the weekly quota. Those who bought before this date will be exempt from the restriction.
MiniMax unveiled the M2.7 model on March 18, demonstrating a path for AI self-evolution. The model is capable of deeply participating in its own training and optimization processes.
In certain research and development scenarios, M2.7 can handle about 30% to 50% of the workload. It achieved an approximate 30% performance improvement on internal evaluation sets.
The model scored 56.22% on the SWE-Pro benchmark, which covers multiple programming languages. The result nearly matches GPT-5.3-Codex and is close to Anthropic's Claude Opus.
In the professional office sector, its GDPval-AA ELO score trails only Claude Opus 4.6 and GPT-5.4, surpassing GPT-5.3, the company claimed.
M2.7 can autonomously build complex Agent Harnesses to complete highly complex productivity tasks. It has demonstrated end-to-end project delivery capabilities in real-world software engineering tasks.
Beyond breakthroughs in productivity tools, M2.7 has also improved role-playing stability. This opens up opportunities for interactive entertainment scenarios, according to MiniMax's statement.
