Baidu Upgrades ERNIE Assistant with Multimodal AI in Baidu Search

Details

• Baidu is rolling out an enhanced ERNIE assistant integrated directly within Baidu Search for mainland users starting 15 October 2025.
• The upgraded assistant can generate and understand eight types of content, including text, images, video, audio, 3-D objects, code snippets, music tracks, and podcast scripts.
• One-click “tool cards” allow users to perform tasks such as summarizing PDFs, drafting lesson plans, troubleshooting code, and planning trips without leaving the search interface.
• The assistant uses the ERNIE 5.0 model, trained on proprietary Wenxin data alongside licensed media libraries to help avoid copyright issues.
• A detachable side-panel interface lets creators link different content formats, enabling workflows like turning sketches into videos with music in a single process.
• Enterprise users receive API access with tiered pricing, starting at ¥0.02 per 1,000 tokens for text and ¥0.09 per second for video content generation.
• Baidu boasts text response times below 800 ms and video previews under five seconds at 720p quality, thanks to recent GPU cluster enhancements.
• Expansion to Hong Kong, Singapore, and a global English-language beta is slated for Q1 2026, dependent on regional compliance approvals.

Impact

This upgrade puts Baidu in direct competition with Tencent and Alibaba, whose tools still separate image and audio creation. By embedding multimodal AI into China's leading search engine, Baidu could influence ad spending and user adoption while aligning with new content regulation. The move sets a high standard for speed and integration, likely shaping the future of generative AI across search and online platforms.