Overview
ERNIE 5.5 Flash Preview is a preview foundation model and the first post-trained model built on ERNIE 5.5 Flash Base. Baidu announced it on November 11, 2026. It has 280 billion total parameters, activates 14 billion per token, and supports a 1M-token context. Baidu reports generation speeds above 200 tokens per second and a default output length of 64K tokens. (ernie.baidu.com)
Capabilities and uses
The model is designed for coding, agentic tasks, knowledge work and deep research. Baidu demonstrates software and game development, 3D modeling, research reports, spreadsheets and video-production workflows. It can work with applications and external tools in an agent environment. (cloud.baidu.com)
Limitations
- The model is described as text input to text output. Demonstrated 3D, document and video workflows use an agent environment and external tools; they should not be assumed to be native multimodal input or output. (cloud.baidu.com)
- It is a preview. Baidu's announcement says the weights are planned for release under Apache 2.0; that statement alone does not establish that they have already been released. (ernie.baidu.com)
Official API pricing
Prices below were checked on November 14, 2026. Domestic and international prices apply to different Baidu Qianfan offerings and are not currency conversions of one another.
Mainland China: CNY
Baidu Qianfan lists the following pay-as-you-go online inference prices per million tokens. The promotional prices are 50% off and expire on November 30, 2026 at 23:59:59; Baidu says original prices will resume afterward. The published page does not specify a time zone for the listed hours. (cloud.baidu.com)
| Period | Input: promotional / original | Cache-hit input: promotional / original | Output: promotional / original |
|---|---|---|---|
| Peak, 08:00–22:00 | CNY 1 / 2 | CNY 0.1 / 0.2 | CNY 4 / 8 |
| Off-peak, 22:00–08:00 next day | CNY 0.5 / 1 | CNY 0.05 / 0.1 | CNY 2 / 4 |
The domestic billing document lists no batch-inference price for this version. Sources: Baidu promotional pricing and Qianfan billing documentation. (cloud.baidu.com)
International: USD
Baidu Qianfan International lists pay-as-you-go online inference at USD 0.28 per million input tokens, USD 0.028 per million cache-hit input tokens, and USD 1.13 per million output tokens. The same rates are listed under text generation and deep thinking. Source: Baidu Qianfan International pricing. (intl.cloud.baidu.com)