At Baidu AI Day, the company released Wenxin Assistant Task Engine 2.0, offering advanced agent capabilities for free, enabling multi-step tasks like converting images to Excel, generating interactive reports, and creating 3D simulations. Additionally, Baidu Search upgraded to GUI interactive answers, providing dynamic content across 2,326 K9 topics.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
At AI DAY, Baidu unveiled the Wenxin Assistant Task Engine 2.0, packaging complex task decomposition, routing, and memory capabilities into a free offering, while transforming search results into interactive GUI components. This marks Baidu's strategic shift from 'where is the information' to 'understand and complete'.
At AI DAY, Baidu officially released the Wenxin Assistant Task Engine 2.0 and announced that high-level Agent capabilities are completely free. This move stands out against the backdrop of similar products charging fees. The core of Task Engine 2.0 lies in translating user goals into executable routes, dispatching different models, tools, and skills through 'task planning' and 'intelligent routing', and using long-context understanding and memory retrieval to avoid 'amnesia' in long tasks.
In testing, the Wenxin Assistant converted four screenshots of Shanghai secondary vocational school independent enrollment admission score lines into Excel, taking 2.5 minutes, organizing 86 major records covering 38 schools, and completing cell merging and formatting. In another cross-platform task, it logged into Baidu Netdisk and Xueqiu, combined with CSRC public information, and generated an interactive HTML report with industry classification, visual charts, and case expansion features.
Baidu Search's results pages are no longer just static content but have been upgraded to 'GUI interactive answers'. For example, typhoon structure diagrams label technical terms on clouds, and industrial revolution timelines allow dragging to see map changes. Such components have reached 2,326, covering high-frequency K9 math and science topics suitable for interactive presentation.
This change means search is shifting from 'information retrieval' to 'understanding and operation', allowing users to drag parameters and view dynamic effects directly on the results page, lowering the barrier to understanding abstract knowledge.
The workflow of Task Engine 2.0 includes 'task planning' and 'intelligent routing'. The planning phase breaks down user goals into steps, and the routing phase selects appropriate models and tools based on the current step, similar to an AI project team with the task engine as project manager. During execution, if results don't meet requirements, the engine continues to check and adjust.
Long-context understanding and memory retrieval capabilities solve the problem of AI forgetting in long tasks, ensuring multi-turn tasks stay on track. These capabilities have been validated in evaluations: in PinchBench v2, the Wenxin Assistant Task Agent achieved a 94.6% overall success rate and 94.4% average score, ranking first; in SuperCLUE XClaw, it scored 97.62, ranking first, with a perfect 100 in memory.
Baidu explicitly stated that the Wenxin Assistant Task Engine 2.0 is completely free, with no paywall and no watered-down version. This contrasts with the current trend of high-level Agents charging fees. Reports note that complex tasks involve significant computing power and tool calls, making costs high, but Baidu chooses to let users access the full capabilities directly.
This strategy may aim to expand the user base, turning high-level Agents from tools for a few professionals into mass productivity. The report comments that this continues Baidu's mission of 'shortening the distance between people and information', further shortening the distance between people and understanding.
This article is based on QbitAI's hands-on report; all feature descriptions and evaluation data come from that report. Evaluation data (PinchBench v2, SuperCLUE XClaw) are third-party, but the report does not provide original report links, so they are marked as source claims. Hands-on results (e.g., 2.5 minutes to convert to Excel, 86 records) are from the journalist's personal testing, but no reproducible raw data is provided, also treated as source claims.
Baidu is pushing high-level Agent capabilities to the masses by free-opening Task Engine 2.0 and launching GUI interactive answers, while backing its technical strength with evaluation scores. However, the sustainability of the free strategy, the independence of evaluation data, and actual user experience still need further verification.
Primary report
Primary source