Recently, the OceanBase team's Data Agent solution topped the international Data Agent Benchmark (DAB), achieving a 90.62% accuracy rate and ranking first. It became the first submission to break through the 90% accuracy threshold on this list. Notably, this achievement was accomplished by OceanBase, a domestic database, using the domestic large model GLM-5.2, surpassing multiple Data Agent solutions built on overseas models such as GPT, Claude Opus, and Claude Fable. The internal code name for this submission is Scout, and its capabilities will be integrated into OceanBase DataPilot.
(Figure caption: OceanBase tops the international Data Agent list, becoming the first submission to achieve an accuracy rate exceeding 90%)
DAB was jointly launched by UC Berkeley EPIC Data Lab and Hasura PromptQL. The evaluation covers multiple fields including internet/local life, finance stocks, biomedical, intellectual property, corporate operations, government/public administration, media/entertainment, and involves various databases such as PostgreSQL, MongoDB, SQLite, and DuckDB.
Unlike traditional Text-to-SQL, which mainly tests whether AI can convert natural language into SQL, DAB focuses on whether AI can find the correct answer from complex, scattered, and diverse data after entering a real data environment. In a complete task, the Data Agent needs to understand the data, select it, plan the analysis path, perform the query and calculation, and verify the results. This tests the full capability from "understanding the data" to "getting the answer."
Therefore, DAB tests not only the underlying model but also the comprehensive ability of "model + Agent + data system." The model is responsible for understanding and reasoning, the Agent for planning and execution, and the data system for supporting data discovery, correlation computation, and result verification. Whether these three components can work together directly affects whether AI can truly utilize enterprise data.
The OceanBase submission for this evaluation was built around this capability. The system uses DataLens to build a data profile, identifying fields and data relationships. Then, based on the complexity of the task, it plans the execution path, completes data selection, filtering, association, and calculation. After obtaining the results, it checks the calculation process and results through evidence tracking and answer validation, adjusting the plan and re-verifying when issues are found, forming a "data understanding - planning and execution - verification and repair" closed loop.
This explains the value of the 90.62% score: it is not just proving that domestic models can perform data queries, but verifying that a combination of domestic databases and models can support complex data analysis tasks. Under the use of GLM-5.2 as the underlying model, the OceanBase solution ultimately surpassed multiple Data Agent solutions built using overseas models such as GPT, Claude Opus, and Claude Fable.
For OceanBase, this is also a new validation of its AI capabilities.
Previously, databases mainly handled data storage, querying, and processing. With AI Agents becoming new data users, the problems databases need to solve are changing: AI not only needs to access data, but also understand, associate, and analyze it, and verify whether the results are reliable. Therefore, the data infrastructure in the AI era is shifting from "delivering data to AI" to "helping AI use data effectively."
This is the direction of OceanBase's evolution from a database to an AI data platform. OceanBase DataPilot is the product direction targeting AI data analysis scenarios. This DAB evaluation was submitted under the internal code name Scout, and the relevant technical capabilities will be integrated into the DataPilot product, further handling data understanding, task planning, analysis execution, and result verification, enabling AI to move from "calling data" to "using data to complete tasks."
From a database to an AI data platform, the changes are not just in product form but also in the relationship between databases and AI: previously, databases were responsible for storing, computing, and retrieving data, but now they also need to help AI understand, organize, and analyze data. This top ranking in DAB is a validation of OceanBase extending its database capabilities into the AI data analysis stage, indicating that databases are evolving from the data infrastructure of AI to the data working engine of AI.
