At Baidu Create Developer Conference, Baidu founder Robin Li unveiled the new-generation Ernie X1Turbo large language model. This model boasts significantly improved performance while maintaining a highly competitive price: only 1 yuan for one million input tokens and 4 yuan for one million output tokens, a mere 25% of the cost of its competitor, DeepSeek-R1.
According to the introduction, Ernie X1Turbo is built upon the 4.5Turbo deep thinking model. It not only offers superior performance but also shows significant enhancements in chain of thought reasoning, question answering, content creation, logical reasoning, tool usage, and multi-modal capabilities, outperforming both DeepSeek R1 and the latest version of DeepSeek V3.
Furthermore, Baidu simultaneously launched Ernie 4.5Turbo, further strengthening its multi-modal capabilities. Compared to Ernie 4.5, the new version is faster and significantly cheaper, with an 80% price reduction. Ernie 4.5Turbo costs only 0.8 yuan for one million input tokens and 3.2 yuan for one million output tokens, just 40% of the cost of DeepSeek-V3. Even more noteworthy is Ernie 4.5Turbo's outstanding performance in various benchmark tests, achieving an average score of 77.68, surpassing GPT-4o's score of 72.76.
Baidu's release of these two highly cost-effective and high-performance large language models will undoubtedly intensify competition in China's AI large model market. They are expected to significantly impact the existing market landscape, particularly in price-sensitive applications and those demanding peak performance.
