Guolian Minsheng Securities: The era of large models in the Agent system is evolving towards "autonomous employees". Recommended to pay attention to MiniMax-WP (00100) and KNOWLEDGE ATLAS (02513).

Guolian Minsheng Securities released a research report stating that in the era of Agents, large models are evolving from "chat tools" to "autonomous employees". Large model manufacturers that master core algorithms and industry interfaces are expected to benefit deeply from the intelligence dividend of everything becoming smart. It is recommended to pay attention to the "twin stars of large models" MiniMax-WP (00100) and KNOWLEDGE ATLAS (02513), which have successfully listed earlier this year, as the "brain" of the native Agent ecosystem, making the company highly scarce. Key Points from Guolian Minsheng Securities: Events As of February 2, 2026, Clawdbot has received over 130,000 stars on the code hosting platform GitHub, with over 2 million cumulative visits to its official website, making it one of the fastest-growing open-source technology projects recently. Additionally, the recent emergence of "AI-only communities" like Moltbook, which has gathered millions of agent accounts in a very short time, naturally correspond to higher request densities and more frequent API triggers. The most direct external variable is the stepwise increase in API call frequency and token throughput. At the recommendation of Clawdbot founder Peter Steinberger, the Chinese AI unicorn MiniMax, specializing in long text and logical reasoning with its M2.1 model, has gained successful popularity. Importance of unit cost per model In the traditional conversation paradigm, only a few model calls are required for each interaction; however, in the workflow paradigm, a task often spans multiple stages such as planning, retrieval, tool invocation, verification and correction, and external system writing. This leads to a doubling of the model call frequency, context length, and complexity of intermediate information. Multi-step reasoning and multi-round tool calls naturally bring about "multi-round context," while retries and self-corrections will generate additional invalid tokens. Compared to basic chatting, agent services for complex tasks may consume several times more tokens. Therefore, the "unit cost of the model unit output" becomes the "life line" of whether Agent class products can be scaled down because in task execution, multi-round reasoning and tool coordination will linearly amplify costs. For this reason, the founder of Clawdbot recommended MiniMax, as its M2.1 model features "efficiency and cost advantages, strong long text abilities, and reasoning and programming capabilities" that meet the current needs of many users. Efficiency and cost effectiveness The M2.1 model aims to address the high token cost pain points faced by developers in automated programming with ultimate cost advantages, priced at approximately 8% of Claude Sonnet's. Additionally, the Coding Plan innovatively introduces a high-frequency refresh mechanism of "resetting quotas every 5 hours," breaking the industry's common daily or monthly limit quota mode, unleashing productivity in high-frequency and heavy development scenarios. In terms of billing, unlike the token-based billing logic commonly used by underlying large model manufacturers (Pay-as-you-go), the company instead adopts a layered monthly subscription system. Strong long text capabilities In real workflows, continuously evolving contexts typically include tool calls, historical information, retrieval snippets, constraints, and more. The long text capabilities of M2.1 make it more suitable for "continuous memory," reading longer documents, accommodating more intermediate results, and reducing logical fractures caused by truncation. Reasoning and programming capabilities In Clawdbot, a product that emphasizes automated execution and correction loop, the model is used for writing code, modifying code, making judgments, and performing verification. M2.1's "sufficient and cost-effective" reasoning and programming capabilities make it the best choice for being put into production systems and being frequently called. In the era of Agents, while "who is smarter" is undoubtedly important, it is even more crucial to be able to "turn strong capabilities into high-frequency usable productivity at a lower cost." The industry believes this is the advantage of MiniMax. Multimodal and "visual execution" taking the forefront After entering office and production scenes, Agents no longer primarily receive input from pure text, but a large amount of visual information such as screenshots, PDFs, tables, charts, interface elements, etc. In Clawdbot's "executable" workflow, users not only input structured text but also come with screenshots, web interfaces, error pop-ups, tables/charts, or PDF pages. MiniMax's multimodal capabilities help Agents better understand interfaces, extract key information, output executable steps/code, check accuracy using screenshots, and make corrections. This allows Clawdbot to perform "visually-driven automation," such as automatically filling in table fields after recognizing them, locating reasons and changing scripts from error screenshots, extracting data from charts and writing it into reports, comparing before and after screenshots to confirm if tasks have really been completed, and so on. With its multimodal capabilities, MiniMax can better complete the service loop, reduce manual reiterations, quickly correct errors, and achieve stronger deliverability. Risk Warning: Technological roadmaps have uncertainties; industry competition is intensifying.