Baidu Catching Up in the AI Race with New ERNIE Models

ERNIE 4.5 outperforms in multiple benchmarks while costing just 1% of GPT-4.5’s price. X1 matches the capabilities of DeepSeek-R1 at half the price.

Baidu has once again solidified its position at the forefront of artificial intelligence with the launch of two groundbreaking models: ERNIE 4.5, its latest foundation model, and ERNIE X1, a deep-thinking reasoning model.

These innovations mark a significant step in China's AI development, showcasing advanced multimodal capabilities and deep reasoning at an unprecedented scale.

ERNIE Bot Goes Free Ahead of Schedule

While the release of ERNIE 4.5 and ERNIE X1 was planned, Baidu has made ERNIE Bot—powered by all ERNIE models—completely free to the public ahead of schedule. Initially set for full access on April 1, this accelerated rollout allows users to interact with all models immediately through the official ERNIE Bot platform

This strategic shift underscores Baidu’s ambition to democratize AI, allowing individual users to leverage cutting-edge AI tools without cost barriers.

Meanwhile, enterprise users and developers can access ERNIE 4.5 via APIs on Baidu AI Cloud’s MaaS platform Qianfan, with ERNIE X1 expected to follow soon.

The foundation model series that powers various generative AI products, including ERNIE Bot: The generative AI chatbot.

With the addition of ERNIE 4.5 and ERNIE X1, Baidu decided to make ERNIE Bot free for public use earlier than planned, allowing users to experience all available models at no cost.

ERNIE 4.5: A Leap Forward in Multimodal Intelligence

ERNIE 4.5 represents Baidu’s most advanced native multimodal foundation model to date. Designed with joint optimization across text, image, audio, and video comprehension, the model offers remarkable improvements in:

• Language understanding and generation

• Logical reasoning and memory

• Hallucination prevention and coding abilities

Compared to OpenAI’s GPT-4.5, Baidu claims ERNIE 4.5 outperforms in multiple benchmarks while costing just 1% of GPT-4.5’s price. The model’s intelligence extends beyond traditional text-based AI, demonstrating an ability to process and interpret internet memes, satirical cartoons, and other contextual content with high accuracy.

The model’s breakthroughs are powered by key technologies such as:

• FlashMask Dynamic Attention Masking

• Heterogeneous Multimodal Mixture-of-Experts

• Spatiotemporal Representation Compression

• Self-feedback Enhanced Post-Training

With input and output pricing starting at RMB 0.004 per thousand tokens and RMB 0.016 per thousand tokens, ERNIE 4.5 positions itself as a cost-effective yet high-performance AI solution for enterprises and developers.

ERNIE X1: Deep-Thinking Reasoning with Tool Integration