Skip to content
DigIntoAI/Model intelligence
Model intelligence feed

Model intelligence.

Follow the changes. Understand what comes next.

Editor’s pickBaidu
Baidu/ERNIE-Wizshot-ME

Baidu ERNIE-Wizshot-ME: Generating Talking-Head Videos from a Portrait Image and Audio

On November 12, 2026, Baidu introduced ERNIE-Wizshot-ME, a dedicated micro-expression video generation model listed on the Baidu AI Cloud Qianfan product portal and API reference documentation (with…

Latest stories04

Newest published
01

Baidu ERNIE-Wizshot-ME: Generating Talking-Head Videos from a Portrait Image and Audio

On November 12, 2026, Baidu introduced ERNIE-Wizshot-ME, a dedicated micro-expression video generation model listed on the Baidu AI Cloud Qianfan product portal and API reference documentation (with…

ERNIE-Wizshot-MERead story
02

Baidu Releases ERNIE 5.5 Flash Preview as ernie-5.5-flash-preview, the First Post-Trained Intermediate Version on Flash Base, Now on Qianfan With a 1M Context

On November 11, 2026 Baidu released ERNIE 5.5 Flash Preview (ID ernie-5.5-flash-preview), the first post-trained version of earlier Flash Base, with 1M context. An intermediate checkpoint: training is not finished; an official release will improve on it. The API is live on Qianfan for text generation and deep thinking. Official prices use yuan per thousand tokens (doc 2026-11-10), with promo and list rates for 8:00–22:00 and 22:00–next day 8:00; no timezone labeled. Weights will be open-sourced under Apache 2.0 soon; not public yet. Official benchmark charts and a full results table are out.

ERNIE 5.5 Flash PreviewRead story
03

What It Takes to Turn ERNIE-5.5-Flash-Base Into a Useful Assistant

Baidu's ERNIE-5.5-Flash-Base offers a starting point for teams building their own language models. Turning the weights into a service involves choosing a task, preparing suitable data and measuring…

ERNIE-5.5-Flash-BaseRead story
04

Baidu Open-Sources Unlimited OCR 2 for Long Documents

On October 27, 2026, Baidu open-sourced Unlimited OCR 2 under MIT on Hugging Face and GitHub—about 3.3B parameters. Sparse Reference Sliding Window Attention (SR-SWA) compresses ultra-long text-visual attention into a fixed-size set of working references so per-step latency stays nearly constant as documents grow. Baidu reports SOTA on MPDocBench and MarathonDocBench; those scores are vendor-reported. Multi-page parsing can run in one request via infer_multi, with Transformers for quick tests and SGLang for serving.

Unlimited OCR 2Read story