Baidu launched its newest artificial intelligence models, Wenxin Big Model 4.5 and Wenxin Big Model X1, intensifying China’s AI battle. The company made the announcement on Sunday, confirming that both models are now live on the Wenxin Yiyan platform and available for free use.
Baidu also confirmed that Wenxin Big Model 4.5 is already accessible on Baidu Smart Cloud’s Qianfan Big Model Platform, allowing developers and enterprises to integrate its API. Meanwhile, Wenxin Big Model X1 is set to follow soon, and the firm is also embedding the models into Baidu Search, Wenxiaoyan App, and other products to expand their reach, per the announcement.
Baidu upgrades AI with multimodal capabilities
Baidu claims that Wenxin Big Model 4.5 is a major step forward in multimodal AI, designed to process text, images, audio, and video. The company says the model is built on native multimodal architecture, improving its ability to understand and generate content across multiple formats. According to the Chinese tech company, it has enhanced logic, reasoning, and memory, while also reducing AI hallucinations.
The model has been trained to recognize internet memes, satire, and complex visual data, a feature Baidu says gives it an edge over previous models. It can reportedly analyze a meme, detect mathematical or logical references, and provide detailed explanations. The company says the model excels in both high-level problem solving and understanding emotional context.
To achieve these upgrades, Baidu integrated several technical advancements:
- FlashMask dynamic attention mask, which speeds up processing for large models and improves long-sequence data handling.
- Multimodal heterogeneous expert expansion, designed to optimize how the AI processes different types of input while reducing gradient imbalances.
- Spatial-temporal representation compression, a system Baidu says compresses and organizes data from long videos more efficiently.
- Large-scale data construction with knowledge points, allowing the model to process high-density knowledge and reduce hallucination errors.
- Post-training self-feedback, which uses reinforcement learning to improve stability and alignment with user intent.
According to Baidu, these improvements make Wenxin Big Model 4.5 faster and more efficient, with better handling of complex content. The company says developers can now integrate the API into enterprise applications, enhancing automated workflows, data analysis, and content generation.
Wenxin Big Model X1 pushes deeper AI integration
Alongside its multimodal counterpart, Baidu is introducing Wenxin Big Model X1, an AI model designed for deep thinking and autonomous tool use. The company claims X1 is its most advanced model yet, with a stronger focus on planning, reflection, and continuous learning. Unlike traditional AI models that rely on pre-set responses, X1 can reportedly adapt by learning from real-time interactions.
Baidu says X1 supports advanced search, document Q&A, image analysis, AI-generated artwork, code interpretation, web link reading, and structured data mapping. The model is also integrated with Baidu Academic Search, business intelligence tools, and franchise data analysis, making it a multi-industry AI assistant.
One of the major capabilities Baidu highlights is X1’s ability to rewrite historical texts using AI-driven contextual analysis. The model reportedly applies a structured approach to replacing historical references while maintaining the original text’s style and structure. It also claims X1 can ensure that rewritten content remains faithful to its source, even when adjusted for modern interpretations.
Baidu attributes X1’s capabilities to several key AI innovations:
- Progressive reinforcement learning, allowing the model to continuously improve based on user interactions.
- End-to-end thought chain training, designed to enhance long-form reasoning and complex decision-making.
- Unified reward systems, which Baidu says help fine-tune AI responses by integrating multiple reward mechanisms for optimized learning.
The company says X1 marks a step toward AI models that think more like humans, with increased adaptability, creativity, and reasoning power.
Baidu has made it clear that Wenxin Big Model 4.5 and X1 are not just stand-alone models but part of a broader AI expansion. The company is embedding these models into its own search engine, cloud services, and enterprise platforms, with plans for wider adoption.
The AI race in China has been heating up, with DeepSeek, iFlytek, and other competitors also launching their own models.
Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot
