Baidu launches ERNIE 4.5 Turbo with an 80% price cut

Background

Multimodal means a model can process more than one kind of input, such as text and images. ERNIE 4.5 was Baidu's general-purpose large language model, and ERNIE X1 was its reasoning model. Developers who call these models through an application programming interface (API) are billed by the number of tokens they send and receive, so a price cut changes the running cost of every application built on top of them. The April 2025 event combined the price cut with announcements aimed at developers.^[1]

What happened

On 25 April 2025 Baidu launched ERNIE 4.5 Turbo and ERNIE X1 Turbo at Create 2025 in Wuhan, an event attended by more than 5,000 developers.^[1] ERNIE 4.5 Turbo cost RMB 0.8 per million input tokens and RMB 3.2 per million output tokens. That is 80% less than ERNIE 4.5, and 40% of the price of DeepSeek-V3.^[1]

Baidu said that the multimodal capabilities of ERNIE 4.5 Turbo matched GPT-4.1 and exceeded GPT-4o on several benchmarks.^[1] AI News reported a multimodal average of 77.68 against 72.76 for GPT-4o.^[2]

Key details

At the event Robin Li announced full support for the Model Context Protocol (MCP). He described MCP as a "universal socket" for AI, and said "Applications are where true power lies."^[1] MCP is an open protocol that lets a model connect to external tools and data sources in a standard way, so that one integration can work across many applications.

Other launches at the event included Xinxiang, a multi-agent app, and Huiboxing, a digital-human livestream platform.^[1]

Impact

For developers, the release combined two changes: a much lower price for the ERNIE 4.5 family and a standard way to connect models to tools.^[1]

Price matters most to teams that send large volumes of requests, because the cost of each request is repeated across many calls. A single percentage cut therefore has a large effect at scale, which is why the April announcement was read by developers as much as by general readers.