Baidu is using its latest Ernie releases to make a clear argument: advanced AI models need to become cheaper, faster, and easier to turn into working applications.
At its Create 2025 developer conference in Wuhan, the Chinese tech giant introduced Ernie 4.5 Turbo and Ernie X1 Turbo. Both models are designed for text and image processing, logical reasoning, and lower operating costs than earlier versions.
Lower prices are central to the Ernie update
The most direct change is price. Baidu has set Ernie 4.5 Turbo at 0.8 RMB (about 11 US cents) per million input text characters and 3.2 RMB (about 44 US cents) per million generated characters.
According to Baidu, that is an 80 percent cost reduction compared to the previous version. The company also says Ernie 4.5 Turbo runs faster and makes fewer errors in text generation than earlier versions.
Baidu is positioning the model against major systems from outside China. The company claims Ernie 4.5 Turbo matches GPT-4.1 in text and image processing and surpasses GPT-4o.
The second launch, Ernie X1 Turbo, is a turbo version of the Ernie X1 reasoning model first introduced in mid-March. Baidu reports that Ernie X1 Turbo outperforms Deepseek-R1 and Deepseek-V3.
Pricing for Ernie X1 Turbo is set at 1 RMB (about 14 US cents) per million input characters and 4 RMB (about 55 US cents) per million output characters. Baidu claims this is a quarter of the cost of Deepseek R1, which is already significantly cheaper than Western models.
Why model pricing matters for developers
For developers, pricing is not just a billing detail. It shapes what kinds of AI products are practical to build, test, and scale.
Lower input and output costs can make frequent model calls more realistic inside apps that handle text, images, and reasoning tasks. That matters for products where the model is not used once, but repeatedly across a workflow.
The two new Ernie models also show how Baidu is trying to compete on several fronts at once:
- Performance claims: Baidu says Ernie 4.5 Turbo improves speed and text-generation accuracy over earlier versions.
- Multimodal capability: Both models are designed to process text and images.
- Reasoning: Ernie X1 Turbo is aimed at logical reasoning tasks.
- Cost reduction: Baidu emphasizes lower prices compared with previous Ernie models and rival systems.
- Access: Both models are available for free within Ernie Bot.
The result is a more aggressive package for developers who are choosing between AI model providers. Baidu is not only presenting new model capabilities, but also trying to lower the friction around using them.
Baidu is putting apps at the center
The Create 2025 event carried the theme “Models Lead, APPs Rule.” Baidu CEO Robin Li emphasized the importance of apps, saying that AI models and chips are worthless without good apps.
That framing is important. Baidu is not treating models as the endpoint. The company is tying them to application layers, developer access, and agent-based tools.
The article notes that this view echoes a recent comment from OpenAI's go-to-market manager Adam Goldberg, who said value creation for AI companies takes place across the entire chain.
In practice, Baidu’s announcements suggest a strategy built around both infrastructure and usable products. The models provide the foundation, while apps, agents, and developer platforms turn that foundation into something people can use directly.
Digital avatars and multi-agent tools expand the strategy
Baidu also introduced the Huiboxing platform. The company claims Huiboxing can generate digital avatars from short video clips, with avatars designed to have a realistic appearance and natural voice.
This moves Baidu’s AI strategy beyond chat and reasoning. Digital avatars point toward applications where generated identity, voice, and visual presence matter.
The company also unveiled Xinxiang, a multi-agent app built for complex task resolution. Baidu claims Xinxiang currently supports 200 task types, including knowledge analysis, travel planning, and office work.
Baidu plans to expand Xinxiang to over 100,000 task types and open developer access. The app is available for Android, while an iOS version is in testing.
Xinxiang also has a technical integration angle. Baidu states that it uses an enhanced version of Anthropic's Model Context Protocol (MCP).
The bigger signal from Baidu’s AI launch
Baidu’s latest announcements combine model competition, lower pricing, and a stronger focus on applications. The company is trying to show that Ernie can compete with systems such as Deepseek-R1, Deepseek-V3, GPT-4.1, and GPT-4o while also giving developers more ways to build on top of its AI stack.
The “AI Open Initiative” fits that direction. According to the company, the platform allows developers to market AI agents, mini-programs, and apps.
That makes Baidu’s message fairly clear. The company wants cheaper models to feed a larger app ecosystem, and it wants developers to see Ernie not only as a model family, but as part of a broader AI product platform.
Whether developers adopt that platform will depend on the practical value of the models, the app tools, and the marketplace around them. But the Create 2025 announcements show Baidu pushing hard on price, capability, and application development at the same time.