Model Positioning
ERNIE 4.5 Turbo VL is Baidu's vision-language model for image understanding. Baidu's official model update log states that its image understanding, content creation, translation, and coding capabilities have improved, and its time to first token has been reduced. (ai.baidu.com)
Capabilities and Uses
The model can be used for image-based understanding and question answering, as well as content creation, translation, and coding-related tasks. The officially listed context length is 128K tokens. (ai.baidu.com)
Limitations
Baidu's international model list specifies a maximum input of 123K tokens and a maximum output of 16,384 tokens for this 128K version, with default rate limits of 1,000 requests and 200,000 tokens per minute; actual limits depend on the region of use and service configuration. The model may produce errors, and important results should be verified by a human. (intl.cloud.baidu.com)
Release Date
Baidu's model update log lists this version as newly available on August 7, 2025; this is the date it became available on the Qianfan platform. (ai.baidu.com)
Pricing
Baidu Qianfan's China-site vision understanding service uses pay-as-you-go billing. The following are the original prices in Chinese yuan (CNY), charged per thousand tokens, and apply to ERNIE-4.5-Turbo-VL and its 32K version:
| Inference method | Input | Output |
|---|---|---|
| Online inference | 0.003 yuan/1,000 tokens | 0.009 yuan/1,000 tokens |
| Batch inference | 0.0012 yuan/1,000 tokens | 0.0036 yuan/1,000 tokens |
Prices and actual availability are subject to Baidu's order page. Official source: Baidu Qianfan Model Service Billing; accessed November 18, 2026. (3e97756f-e20a-4497-8bc2-57f45e005c8a.baidu.scalaris-eci.com)
The Baidu international pricing documentation was checked, but the official price for this model could not be retrieved this time, so no amount is listed. (intl.cloud.baidu.com)