How Baidu's Latest AI Models Will Power Innovation
Baidu is sharpening its technology focus with the unveiling of its ERNIE 4.5 Turbo and ERNIE X1 Turbo models, placing cost and capability at the centre of the evolving AI landscape.
At its annual Baidu Create 2025 conference in Wuhan, the company showcases a raft of innovations designed to democratise access to AI and drive practical applications across industries.
As enterprises worldwide demand greater return on investment from AI implementations and developers seek options beyond Western platforms, Baidu is responding with models that offer enhanced capability at a fraction of previous prices.
Speaking to more than 5,000 developers and technology professionals, Baidu's Co-founder and Chief Executive Officer Robin Li, explains: "Our releases aim to empower developers to build the best applications — without having to worry about model capability, costs or development tools."
ERNIE 4.5 Turbo and X1 Turbo: New Benchmarks in Cost and Capability
The two models, ERNIE 4.5 Turbo and ERNIE X1 Turbo, represent Baidu's latest move in a fast-changing LLM (large language model) market.
Both models feature multimodal capabilities, allowing them to process a mixture of text, images and other data types and are priced much lower than previous versions.
ERNIE 4.5 Turbo, aimed at general-purpose applications, comes with input costs at RMB 0.8 per million tokens and output costs at RMB 3.2 per million tokens.
Baidu claims this is just 0.2% of the cost of GPT-4.5 and 40% of the competing DeepSeek V3 model.
In addition to being cheaper, the model also improves logical reasoning, coding abilities and reduces hallucination rates, which are errors AI models make by generating false information.
Meanwhile, ERNIE X1 Turbo targets deep thinking and chain-of-thought reasoning — a process where the AI model works through problems step-by-step instead of offering immediate answers.
Priced at RMB 1 per million tokens for input and RMB 4 per million tokens for output, this model cuts costs by half compared to its predecessor.
Both models are freely accessible via ERNIE Bot, reflecting Baidu’s goal to lower barriers in the AI sector.
Robin says: "Multimodality will become a standard feature of future foundation models. The market for pure text-based models will shrink, while the market for multimodal models will continue to grow."
Innovation beyond models: digital humans and multi-agent collaboration
Beyond LLMs, Baidu is pushing innovation through its latest AI applications, including the development of digital humans and collaborative agent systems.
Among the standout innovations is the Huiboxing platform, which enables users to create personalised digital avatars from videos as short as two minutes.
Robin describes AI digital humans as "one of the most exciting breakthrough applications in 2025".
Further pushing the boundaries of multi-agent technology, Baidu introduced Xinxiang, described as a "general super agent" platform.
Xinxiang uses multiple AI agents working together to handle complex user tasks.
It currently supports 200 different task types, including knowledge analysis, travel planning, education and office workflows, with ambitions to expand to more than 100,000.
Robin highlights how AI can make programming capabilities accessible to a global audience: "There are nearly 30 million programmers and 8 billion people globally.
"As technologies become more accessible, everyone can possess the capabilities of a programmer."
Cangzhou OS also made its debut at the conference.
Jointly developed by Baidu Wenku and Baidu Drive, this content operating system powers AI Note, the industry's first multimodal AI note-taking tool.
AI Note allows users to generate notes, mind maps and quizzes directly from educational videos.
Baidu Wenku now counts 40 million paying users and 97 million monthly active users, while Baidu Drive's AI features have surpassed 80 million monthly active users.
Open platforms and the future of AI applications
Building on its commitment to technology development and AI integration, Baidu announced the launch of the AI Open Initiative via the Baidu Search Open Platform.
This initiative provides developers with opportunities for traffic generation, monetisation and access to AI services.
Another major announcement is Baidu’s support for the Model Context Protocol (MCP), a standardised interface designed to connect external services with LLMs.
Robin compares MCP to a "universal socket" for AI, demonstrating its integration across Baidu's platforms, including the Qianfan Foundation Model Platform, Baidu Search, ecommerce services and Baidu Drive.
In a bid to encourage more innovation, Baidu unveiled the third ERNIE Cup Innovation Challenge, doubling its prize pool to RMB 70 million.
The company also commits to training an additional 10 million AI professionals over the next five years.
The conference's theme, ‘Models Lead, APPs Rule’, captures the company's focus on practical technology adoption.
Covering areas such as agent-based AI systems and DeepSeek technology, the event underlines Baidu’s vision for the future.
Robin concludes: "Without practical applications, neither advanced chips nor sophisticated models hold value.
"While there will be a multitude of models, it is the applications that will truly dominate in the future. Applications are where true power lies."
Explore the latest edition of Technology Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.
Discover all our upcoming events and secure your tickets today.
Technology Magazine is a BizClik brand
