Baidu's strategic planning Agent 2.0 takes the global lead, outperforming similar agents in high-difficulty tasks

Baidu Smart Cloud's enterprise-level algorithm optimization intelligent body "Baidu Famo Agent 2.0" has recently topped the machine learning engineering benchmark test MLE-Bench again, setting a new state-of-the-art (SOTA) record. Baidu stated that the product will be officially released at the Create 2026 Baidu AI Developer Conference in May this year. MLE-Bench was established by OpenAI and includes 75 real engineering problems sourced from Kaggle competitions, testing AI's end-to-end capabilities across the entire process of model training, data preparation, and experimental execution. Famo first topped the list in October last year. Leading Win Rate in High-Difficulty Tasks: Evolutionary Strategies and Long-Term Memory Upgrades The latest evaluation shows that under unified operating standards, Baidu Famo 2.0 significantly leads in overall win rates on "high-difficulty" tasks, surpassing similar intelligent bodies equipped with mainstream large models such as Claude-Opus-4.6. Baidu stated that Famo 2.0 has undergone upgrades in evolutionary strategies, long-term memory mechanisms, and underlying infrastructure, including enhanced evolutionary strategy support for multi-path parallel exploration and backtracking adjustments, as well as long-term memory mechanisms to help maintain logical consistency in long-chain tasks. Baidu noted that relying on a full-stack AI cloud infrastructure, the efficiency of algorithm evolution and iteration has significantly improved. Additionally, business personnel without an algorithm background can initiate requests using natural language and data files, with the system automatically outputting interpretable and interactive decision-making solutions. Covering Automotive, Finance, Energy, and Other Fields: Research Scenarios Supporting Aerospace and Disaster Prediction Baidu claims that since its launch, Baidu Famo has attracted thousands of enterprises, covering core fields such as retail, finance, manufacturing, energy, and transportation. In automotive manufacturing, Alter Titan has collaborated with Famo to develop the Yufeng intelligent prediction system, reducing the single wind resistance verification time from 10 hours to several minutes, with an average reduction of 25% in the overall vehicle development cycle. In financial risk control, Citic Baixin Bank has introduced Famo to mine risk features 24/7, improving efficiency by 100% and increasing the risk differentiation of risk control models by 2.41%. In energy infrastructure, China Energy Construction Guangdong Institute used Famo to solve the layout problem of offshore wind power cable trays, saving nearly a week of construction time. In traffic signal control, after introducing the Famo signal control platform, the average delay per vehicle in Ordos Yijinhuoluo Banner was reduced by 18%, and peak travel time was reduced by over 50%. The team from Beijing University of Technology introduced Famo into the design experiment of micro gas chromatography columns for the Chinese space station, significantly improving separation efficiency; the team from Tianjin University used it for disaster prediction model optimization, compressing research exploration originally measured in "weeks" to results within 6 hours. Recently, Baidu Smart Cloud also open-sourced the Famou for Science project, building virtual research teams based on a multi-agent collaborative model to support the automated advancement of long-term research tasks