Baidu Unveils New Language Models and AI Applications

Introduction of New Language Models

Baidu, a leading technology company in China, has recently unveiled two advanced language models during its Create 2025 developer conference held in Wuhan. The new models, named Ernie 4.5 Turbo and Ernie X1 Turbo, are designed to handle both text and image inputs, as well as logical reasoning tasks. Notably, these models are positioned at a lower price point compared to their previous iterations.

Key Features of Ernie 4.5 Turbo

Enhanced Performance: Baidu touts that Ernie 4.5 Turbo operates more swiftly and generates fewer errors compared to earlier versions. The model is reported to perform at par with OpenAI’s models, specifically GPT-4.1, and even outperforms GPT-4o in some tasks.
Affordability: The pricing for Ernie 4.5 Turbo is significantly reduced, costing approximately 0.8 RMB (around 11 US cents) for every million input text characters, and 3.2 RMB (about 44 US cents) for each million characters generated. This indicates an 80% reduction in costs compared to the older version.

Summary of Pricing and Performance

Performance: Comparable to GPT-4.1 and GPT-4o.
Input Cost: 0.8 RMB per million text characters.
Output Cost: 3.2 RMB per million generated characters.

Launch of Ernie X1 Turbo

In addition to Ernie 4.5 Turbo, Baidu has also introduced Ernie X1 Turbo, an upgraded version of its reasoning model. This model aims to outperform competitors such as Deepseek-R1 and Deepseek-V3.

Cost-Effectiveness of Ernie X1 Turbo

Competitive Pricing: The pricing for Ernie X1 Turbo is set at 1 RMB (around 14 US cents) per million input characters and 4 RMB (about 55 US cents) for million output characters. Baidu claims this represents only a quarter of the cost of Deepseek R1, which is already cheaper than many Western counterparts.
Availability: Both Ernie models are accessible for free within the Ernie Bot platform, encouraging broader usage and experimentation.

New AI Applications and Innovations

Baidu is not only refining its language models but is also expanding into new realms of AI technology. The company presented the Huiboxing platform, which can create digital avatars from short video clips. These avatars aim to appear realistic and exhibit natural speech.

Introduction of Xinxiang Multi-Agent App

Another exciting development is the launch of Xinxiang, a multi-agent application designed for complex tasks, which currently supports around 200 task types such as knowledge analysis and travel planning. There are plans to expand Xinxiang’s capabilities to over 100,000 task types and allow developers access to this platform.

Technical Integration and Developer Engagement

Baidu notes that Xinxiang incorporates an improved version of Anthropic’s Model Context Protocol (MCP) for its technical integration. The company aims to enhance its collaboration with developers through the AI Open Initiative, which facilitates the marketing of AI agents, mini-programs, and apps.

Emphasis on AI Applications

At the Create 2025 event, Robin Li, CEO of Baidu, emphasized that while AI models and chips are crucial, their real value lies in practical applications. This sentiment mirrors recent remarks by industry leaders, underscoring the importance of app development in realizing the full potential of AI technologies.

Future Scope of Baidu’s AI Journey

Baidu’s continuous innovation in AI demonstrates its commitment to advancing technology while keeping costs competitive. As their language models and applications evolve, they position themselves firmly within the global AI landscape, challenging both local and international contenders. The company’s proactive approach signals a pivotal shift in how AI can be integrated into everyday tasks and applications.

Please follow and like us: