Back home

文心闪电深度思考模型 / ERNIE-Light-DT-1.0

x+web 0.93
Visit source

Intro

文心闪电深度思考模型 / ERNIE-Light-DT-1.0 is one of two Wenxin application models Baidu formally launched at the opening of Baidu World 2026 on 12 November 2026. National Business Daily (2026-11-12 11:09:13) reported Robin Li saying Baidu would keep optimizing models for applications; both models went live on Baidu AI Cloud’s Qianfan platform with API access for enterprise customers and partners.

Qianfan’s model list (updated 2026-11-11) lists ERNIE-Light-DT-1.0 under both text generation and deep thinking. The endpoint / model id is ernie-light-dt-1.0, with a 1M-token context, 1M max input, 128k max output, and default rate limits of RPM = 60 and TPM = 1,000,000. It sits in the ERNIE flagship table beside ERNIE-5.5-Flash-Preview and ERNIE 5.1. The public table does not use the Chinese announcement name as a separate column; cite both names together.

The deep-thinking guide (updated 2026-11-11) includes ernie-light-dt-1.0 among models that can turn thinking on or off with the thinking parameter, and says it supports reasoning_effort (default max; allowed values low, high, max). The chat endpoint is https://qianfan.baidubce.com/v2/chat/completions.

Model pricing (updated 2026-11-11) lists online inference for ERNIE-Light-DT / ERNIE-Light-DT-1.0 at 0.008 yuan per thousand input tokens, 0.0008 yuan per thousand cache-hit tokens, and 0.028 yuan per thousand output tokens. Recheck that page for the latest figures.

Key points:

Explain

Why it matters

This is a named application-oriented reasoning model announced at Baidu World on 12 November 2026 and already present on Qianfan’s official model list and pricing page, with a clear endpoint, context length and list price. That is enough for a standalone radar entry.

This page only restates the keynote coverage and Qianfan docs. Public materials do not give parameter counts or independent scores.

Limits

The Chinese announcement name and the English table name should be cited together. No parameter count or benchmark is in the docs reviewed here. Prices and rate limits can change. Do not describe it as free for every developer, or as the same model as ERNIE 5.5 Flash.

Usage

  1. Compare the keynote report https://www.nbd.com.cn/articles/2026-11-12/4645732.html , the model list https://cloud.baidu.com/doc/qianfan-docs/s/7m95lyy43 , the thinking guide https://cloud.baidu.com/doc/qianfan-docs/s/Wm95lyynv , and pricing https://cloud.baidu.com/doc/qianfan/s/wmh4sv6ya ;
  2. Call Qianfan’s chat completions API with model ernie-light-dt-1.0; use thinking to toggle reasoning and reasoning_effort (low / high / max, default max) as documented;
  3. When quoting prices, context length or rate limits, cite the 2026-11-11 Qianfan docs and re-check upstream;
  4. Do not confuse 文心闪电深度思考模型 / ERNIE-Light-DT-1.0 with ERNIE 5.5 Flash or with the same-day ERNIE-Wizshot-ME video model.

Advertisement