On March 2, 2026, Akshay Kothari, co-founder of Notion, the world's top collaborative office platform, announced a major initiative - Notion Custom Agents officially introduced the open-source weight model MiniMax M2.5 developed by MiniMax, and opened it to users worldwide as an experimental feature.

This news is like a bombshell, causing a huge uproar in the technology industry. It not only breaks the long-standing monopoly of Notion by top closed source models such as Claude and GPT, but also marks the official entry of domestic large models into the core field of global mainstream productivity tools.

Breaking the monopoly of closed source, domestic big models emerge

For a long time, Notion has relied on closed source models such as Claude and GPT for model selection. Although these models have excellent performance, they also have certain limitations. The range of choices for users is relatively narrow, and the cost of use is high.

The emergence of MiniMax M2.5 has brought a brand new choice to Notion's over 100 million users. In Notion's model selection dropdown menu, the MiniMax M2.5 is now on par with top models such as Claude Sonnet 4.6 and Gemini 3 Pro, indicating that domestic models are starting to compete with closed source giants on the global stage.

Ultimate performance and "cabbage price" cost, the core advantages of MiniMax M2.5

The key reason why MiniMax M2.5 quickly gained Notion's favor is its ultimate performance and ultra-high cost-effectiveness.

For simple but high-volume daily tasks such as document organization, schedule synchronization, and automation, the calling cost of MiniMax M2.5 is much lower than that of closed source models. This makes it a "productivity necessity" in the AI industry, especially suitable for users and enterprises who are cost sensitive but require a large amount of AI assisted work.

For example, some small startups or start-up teams, with limited resources, can greatly reduce operating costs and improve work efficiency by using MiniMax M2.5.

In terms of technical architecture, MiniMax M2.5 adopts the RLAIF framework optimized for Agent Workflow. This framework makes it extremely robust in handling complex task disassembly and multi-step execution. Whether facing complex project planning or tedious data analysis, MiniMax M2.5 can complete tasks in an orderly manner, providing users with accurate and efficient solutions.

Technical endorsement: Strong performance, vertical office optimization

Since its release in February 2026, the MiniMax M2.5 has attracted widespread attention in Silicon Valley. Its outstanding performance in multiple aspects has laid a solid foundation for its entry into Notion.

In terms of performance, MiniMax M2.5 performs well in hard core programming and tool calling rankings such as SWE Bench, even surpassing some closed source flagship models. Its inference speed is as high as 100 tokens/second, which can quickly respond to user needs and provide users with a smooth user experience. Whether it is coding, algorithm optimization, or debugging complex systems, MiniMax M2.5 can demonstrate its powerful capabilities.

In terms of vertical office optimization, the MiniMax M2.5 is particularly innovative. The model deeply simulated real office scenarios such as Excel financial modeling, Word deep writing, and PPT logic generation during the training process. This perfectly aligns with Notion's "all-in-one workbench" attribute, allowing users to leverage the powerful capabilities of MiniMax M2.5 to improve work quality and efficiency when using Notion for various office tasks.

For example, when conducting financial analysis, MiniMax M2.5 can quickly generate accurate financial statements and data analysis results; When writing documents, it can provide rich creativity and professional advice to help users create high-quality documents.

Industry Observation: Notion's Strategic Transformation from "Omnipotent Brain" to "Professional Division of Labor"

Akshay Kothari's introduction of open-source models reflects Notion's core strategy for 2026- Model Agnosticism. Under this strategy, users are no longer locked into the hands of a single supplier, but can freely switch models based on task requirements.

For example, when users pursue deep literary creation, they can choose Claude Opus 4.6, whose powerful language generation ability can meet users' creative needs for literary works; When users engage in large-scale data cleaning and logical execution, MiniMax M2.5 is a better choice, as its efficient processing and precise logical analysis capabilities can help users quickly complete data tasks.

This' hybrid 'mode will help Notion further consolidate its position as the' personal and enterprise digital brain 'in the AGI era. It can not only meet the diverse needs of different users, but also improve the flexibility and adaptability of the platform, enabling Notion to maintain a leading position in the fierce market competition.