Baidu’s Ernie 4.5 & X1: Redefining AI with Multimodal Power

The competition among global AI providers continues to intensify as companies try to deliver more capable foundation models at increasingly competitive price points.

This trend is especially evident in the Chinese market, where domestic technology firms are working to establish technological parity with Western counterparts, while offering pricing structures that could potentially reshape market dynamics.

DeepSeek recently exemplified this trend when it disrupted the market by introducing its R1 model, which reportedly offers performance comparable to leading proprietary models at just a fraction of the cost – encouraging industry players to reassess their value propositions and pricing strategies.

Now, Baidu has launched two new foundation models that expand its AI capabilities: Ernie 4.5, a native multimodal foundation model and Ernie X1, a deep-thinking reasoning model – both now freely accessible to individual users through the Ernie Bot platform.

This release comes as industry observers note that pricing strategies may become as important as technical capabilities in determining market adoption.

As a result, Baidu's approach of offering consumer access for free while maintaining enterprise pricing at a fraction of competitors' rates signals a potential shift in how AI capabilities might be monetised going forward.

Baidu is offering free access to consumers earlier than its previously announced April 1 date – and enterprise users and developers can access Ernie 4.5 via APIs on Baidu AI Cloud's platform Qianfan, with Ernie X1 set to join the offering soon.

Furthermore, Baidu positions these models as direct competitors to OpenAI's GPT series.

Baidu Ernie 4.5: bringing native multimodal capabilities at competitive price point

Ernie 4.5 is Baidu's latest generation of foundation models developed internally by the company.

The model utilises joint modelling of multiple data types to achieve collaborative optimisation, resulting in improved multimodal comprehension capabilities.

It also incorporates several technical advances that contribute to its performance, including FlashMask Dynamic Attention Masking – a technique that helps the model focus on relevant parts of input data – and Heterogeneous Multimodal Mixture-of-Experts, which allows the system to specialise different components for handling various types of information.

Baidu additionally highlights the model's “high EQ” from its ability to understand internet memes and satirical cartoons as evidence of its contextual awareness – because its multimodal design enables it to process and comprehend text, images, audio and video content in an integrated manner.

Baidu says Ernie 4.5 has “excellent multimodal understanding ability. It has more advanced language ability and its understanding, generation, logic and memory abilities are comprehensively improved.”

Baidu Ernie X1: emphasising deep reasoning capabilities and tool integration

Ernie X1, Baidu's first multimodal deep-thinking reasoning model capable of tool use, focuses on “stronger understanding, planning, reflection and evolution capabilities” and “delivers performance on par with DeepSeek R1 at only half the price,” according to the company.

The model also demonstrates strengths in areas requiring complex reasoning, including knowledge-based questions, literary creation, logical reasoning and mathematical calculations.

It integrates with various tools including advanced search functionality, document question-answering, image understanding and generation, code interpretation and specialised search functions for academic, business and franchise information.

Furthermore, technical innovations supporting X1 include a Progressive Reinforcement Learning Method, which helps the model improve through feedback and End-to-End Training Approach Integrating Chains of Thought and Action, which combines reasoning pathways with actions the model can take.

In the future, Baidu intends to incorporate both Ernie 4.5 and X1 throughout its product ecosystem.

This integration will extend to Baidu Search, the company's internet search engine with over 600 million users, as well as the Wenxiaoyan application and other Baidu offerings.

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