en.Wedoany.com Reported - On June 15, Baidu's DuMate announced the completion of a core engine upgrade. Through continuous optimization of the Harness engine and multiple engineering aspects, this upgrade reduces Token consumption during task execution by 75% while ensuring that the Agent's intelligent capabilities and task execution performance remain unaffected, with corresponding user point consumption also reduced by 75%. For enterprise-level intelligent agent products, Token consumption directly impacts usage costs, task response efficiency, and the feasibility of large-scale deployment. This upgrade advances cost optimization from the model invocation layer to the Agent execution chain layer.
DuMate is a general-purpose intelligent agent product for office and enterprise use launched by Baidu, targeting scenarios such as file processing, browser operations, Office collaboration, information retrieval, task decomposition, and cross-tool execution. Compared to standard Q&A models, the cost pressure for Agent products is more concentrated in the task execution process: it requires repeatedly understanding the screen, invoking tools, reading context, generating operation steps, verifying results, and making corrections after failures. A complex task is often not completed with a single model invocation but through multiple rounds of planning, execution, and verification. Higher Token consumption amplifies user waiting time, platform inference costs, and point consumption.
The key to this core engine upgrade lies in the Harness engine. Harness can be understood as the engineering layer that connects the model, tools, task workflows, and execution environment for the intelligent agent. It determines how the model decomposes tasks, when to invoke tools, how to reuse context, how to compress unnecessary steps, and how to re-plan after task failures. Baidu previously mentioned "engineering orchestration" capabilities in its intelligent cloud product upgrades, emphasizing the use of engineering optimization to improve office task execution success rates and reduce Token consumption. DuMate's 75% reduction in Token consumption indicates that the optimization focus has shifted from single-query efficiency to cost control across the entire Agent task execution path.
Such optimization will be more directly perceptible to users. When enterprise employees use intelligent agents for tasks like spreadsheets, documents, web searches, data organization, and process entry, their primary concerns are whether the task can be completed, whether wait times are acceptable, and whether point consumption is too high. If Token consumption for the same task decreases, cost estimation during trial and deployment phases becomes clearer for enterprises, and individual users are more likely to use Agents for high-frequency office tasks rather than only experimenting with a few complex scenarios. For intelligent agent products, cost reductions often increase usage frequency, leading to more real-world task feedback.
The Agent industry is transitioning from "can it be done" to "can it be completed stably at low cost." Early intelligent agent products primarily showcased capabilities like browser control, file operations, and multi-step planning, but in practice, they often encountered issues such as high costs for long tasks, redundant steps, context bloat, and excessive consumption from failure retries. The value of Harness layer optimization lies in enabling the intelligent agent to reduce unnecessary thinking and repeated invocations, focusing model capabilities on critical judgments, complex reasoning, and result validation. As office, marketing, customer service, R&D, and operations scenarios continue to integrate Agents, execution cost will become a key metric for enterprise procurement and sustained use.
This DuMate upgrade also reflects a shift in the competitive focus of domestic general-purpose intelligent agent products. Model capabilities remain foundational, but product experience increasingly depends on engineering systems, tool orchestration, context management, and cost control. Those who can reduce Token consumption while maintaining task success rates will find it easier to enter high-frequency enterprise scenarios. For Baidu, the DuMate core engine upgrade helps enhance its product competitiveness in the office intelligent agent and enterprise Agent market, while also laying a cost foundation for subsequent local deployment, team collaboration, and industry scenario expansion.
Subsequent observation of actual task performance is still needed. A 75% reduction in Token consumption is a clear cost optimization metric, but enterprise users will also focus on complex task success rates, file processing accuracy, tool invocation stability, privacy security, permission control, and multi-user collaboration capabilities. If DuMate can consistently maintain execution performance in high-frequency office tasks and ensure that cost reductions are genuinely reflected in user points and enterprise budgets, this upgrade will be more than just a parameter adjustment—it will be a validation of engineering capabilities for the large-scale adoption of general-purpose intelligent agents.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
