Baidu powers up Kunlun super-cluster, unveils cutting-edge AI models | A compact guide to large language models pdf download | Large language models tutorial pdf | Llm datasets github | Turtles AI

Baidu has activated a cluster of 30,000 Kunlun P800 chips, supporting the training of advanced AI models. It has launched the Ernie 4.5 Turbo and Ernie X1 Turbo models, with competitive performance compared to GPT-4.5 and DeepSeek R1. Both models are freely accessible via Ernie Bot. Baidu aims to integrate these technologies into its cloud and content platforms.

Key Points:

• The Kunlun P800 cluster enables the training of AI models with hundreds of billions of parameters.
• Ernie 4.5 outperforms GPT-4.5 in multiple multi-modal benchmarks, with reduced operating costs.
• Ernie X1 offers advanced reasoning capabilities, comparable to DeepSeek R1, at half the price.
• Baidu makes Ernie Bot free for individual users, promoting the adoption of its AI technologies.

At its annual developer conference, Baidu announced the activation of a cluster of 30,000 self-developed third-generation Kunlun P800 chips for training large-scale AI models. The cluster can support models with hundreds of billions of parameters, similar to DeepSeek, and allows thousands of customers to train models simultaneously. P800 chips have already been adopted by Chinese banks and internet companies.

In parallel, Baidu unveiled two new AI models: Ernie 4.5 Turbo and Ernie X1 Turbo. Ernie 4.5 is a multimodal model that integrates text, images, audio, and video, outperforming GPT-4.5 in several benchmarks, including C-Eval and CMMLU, with an operating cost of 1% of competitors. Its performance is attributed to technologies such as "FlashMask" and "Heterogeneous Multimodal Mixture-of-Experts." Ernie X1, on the other hand, focuses on deep reasoning, excelling at complex tasks like manuscript writing and image analysis, offering comparable performance to DeepSeek R1 at half the price.

Both models are accessible for free via Ernie Bot, which is available to individual users. Baidu plans to integrate these AI capabilities into its cloud and content platforms, such as Baidu Wenku, and has announced plans to open-source Ernie 4.5 by June 30. These moves come amid growing competition in China’s AI space, with companies like Tencent and Alibaba developing advanced models for practical applications. 

Baidu is solidifying its position in the AI ​​landscape by combining advanced hardware infrastructure with high-performance, accessible AI models, aiming for widespread adoption of its technologies.