Baidu Deploys ERNIE Lightning and ERNIE-Wizshot-ME on Qianfan Platform
AI
Reported by Dr. Kai Wei, Frontier AI Desk
- Baidu introduced ERNIE Lightning, a sparse-attention model for fast, multi-step search tasks.
- ERNIE-Wizshot-ME creates 4-15 second 1080p videos from still images using text, vision and speech inputs.
- Both models are available via APIs on Qianfan, with pricing of 0.7 yuan per video second for Wizshot-ME.
Baidu launched two new ERNIE models for search and micro-expression video on its Qianfan AI Cloud, targeting enterprise users.
«Baidu Puts ERNIE Lightning Reasoning Model and ERNIE-Wizshot-ME Expressive Video Model on Qianfan»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
Baidu announced two application-focused ERNIE models at Baidu World 2026 and made them live on the Baidu AI Cloud Qianfan platform, with APIs open to enterprise customers and partners link. ERNIE Lightning, built for Baidu Search, uses Baidu Sparse Attention (BSA) and a “looping” scheme that pairs a fast-thinking model with a slower judge model. Baidu says Kunlunxin co-optimization lets the model run inference on a single eight-card Kunlunxin server, cutting inference engine cost by more than 30%.
ERNIE-Wizshot-ME targets emotional expression in AI video. It accepts scripts up to 4,000 characters, maps cues across six dimensions and 200 sub-dimensions, and produces 1080p clips lasting four to 15 seconds. Qianfan lists the service at 0.7 yuan per second and claims production costs are 50 % below industry levels, with a fast mode that generates one minute of video in about one minute.
Baidu has not released independent benchmarks, parameter counts, or context-length specifications for either model.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.
