Published on: March 17, 2025 / Updated on: March 17, 2025 – Author: Konrad Wolfenstein
Baidu is attacking with its new AI models Ernie 4.5 and Ernie X1: comparable to Deepseek, but at half the price – Image: Xpert.Digital
Ernie 4.5 & X1: Baidu's new AI models set new standards
Baidu's new AI models Ernie 4.5 and X1: Powerful at half the price
On March 16, 2025, Baidu unveiled two new AI models that could shake up the fiercely competitive Chinese and global AI market. With the entry-level model Ernie 4.5 and the logic-focused Ernie X1, the Chinese tech giant is pursuing an aggressive pricing strategy while promising high performance. Particularly noteworthy is the claim that Ernie X1 will offer the same performance as the successful DeepSeek R1 at half the cost. Furthermore, Baidu plans to release its models as an open-source solution later this year, which could further intensify competition.
Suitable for:
The new Ernie generation: Technical features and capabilities
Ernie 4.5: The multimodal flagship
According to Baidu, Ernie 4.5 is the company's first "natively multimodal" language model and was specifically designed to process and understand various data types such as text, images, audio, and video. This multimodal integration capability represents a significant advancement in AI development and opens up new application possibilities.
The model is said to possess “excellent multimodal understanding capabilities” as well as improved language, generation, logic, and memory functions. A particular feature is its high “Emotional Quotient” (EQ), which is intended to enable it to understand memes and satirical content – a capability previously attributed primarily to Western AI models like ChatGPT.
According to Baidu, five technological innovations underlie the impressive capabilities of Ernie 4.5: “FlashMask” Dynamic Attention Masking for more efficient data processing, Heterogeneous Multimodal Mixture-of-Experts for specialized processing of different content types, Spatiotemporal Representation Compression for optimizing spatial and temporal information, Knowledge-Centric Training Data Construction, and Self-feedback Enhanced Post-Training.
Ernie X1: The Specialized Thinker
Ernie X1 is designed as a “deep-thinking model with multimodal capabilities” and specializes in logical reasoning. It is characterized by “stronger comprehension, planning, reflection, and evolutionary abilities.” As Baidu’s first “thinking model,” Ernie X1 supports a comprehensive suite of tools that extend its functionality.
The technical foundations of Ernie X1 are based on several advanced technologies, including Progressive Reinforcement Learning Method and End-to-End Training Approach. This structured learning approach is designed to help the model improve incrementally and continuously increase its performance.
Compared to the competition: aspiration and reality
Ernie X1 vs. DeepSeek R1
Baidu is aggressively positioning Ernie X1 against DeepSeek R1, one of China's most successful AI models. According to Baidu, Ernie X1 offers "the same performance" as DeepSeek R1, but at "half the cost." This is a remarkable claim, given DeepSeek R1's reputation for its strength in complex tasks.
DeepSeek R1 has achieved impressive results in various benchmarks in the past and directly competes with OpenAI's o1 models. In mathematics benchmarks such as AIME 2024, DeepSeek R1 achieved 79.8%, just ahead of OpenAI o1-1217 at 79.2%. On MATH-500, it even scored 97.3%, slightly above OpenAI o1-1217 at 96.4%. DeepSeek R1 also demonstrates strong performance in programming tasks and general knowledge tests.
Whether Ernie X1 can actually match these performance levels cannot yet be independently verified. Baidu has not yet published any benchmark results for its new thought model.
Ernie 4.5 vs. GPT-4.5
Baidu goes even further with Ernie 4.5, claiming that the model outperforms GPT-4.5 and DeepSeek V3 in several benchmarks. It is said to excel particularly in multimodal tasks and logical reasoning. However, independent tests and comparisons that could confirm these claims are lacking.
Aggressive pricing strategy: The cost advantage in detail
Given the intense competition in China, Baidu is pursuing an aggressive pricing strategy aimed at both private users and businesses.
Free for private use
Both models are available to private users free of charge. Access is via the official ERNIE bot, and even earlier than originally planned. This strategy aims to rapidly expand the user base and catch up with competitors like ByteDance and DeepSeek.
Unbeatable prices for business customers
Baidu offers its models to corporate customers at extremely competitive prices:
- Ernie 4.5: $0.55 per million tokens for input, $2.20 for output
- Ernie X1: $0.28 per million tokens for input, $1.10 for output
For comparison: DeepSeek R1 costs $0.55 to buy and $2.19 to issue per million tokens. Ernie X1 is therefore actually about half as expensive as DeepSeek R1, at least in terms of issuance costs.
Even more dramatic is the price difference compared to GPT-4.5, which reportedly costs $75 per million input tokens and $150 per million output tokens. Ernie 4.5 is therefore only about 1% as expensive as GPT-4.5. This extreme price difference could represent a significant competitive advantage, provided the performance promises are met.
Suitable for:
Open-source strategy and availability
Early start and broad accessibility
Baidu has made its new models available earlier than planned. Originally announced for April, they are already accessible via the ERNIE Bot. For businesses, Ernie 4.5 and X1 are available via APIs on Baidu AI Cloud's MaaS platform, Qianfan.
Open-source plans for June 2025
In a strategic shift, Baidu plans to release its Ernie 4.5 series as an open-source solution starting June 30, 2025. CEO Robin Li, who has long advocated for proprietary models, justifies this move by stating that open source accelerates the adoption of the technology. Whether this also applies to the X1 series is currently unclear, but seems likely, given that DeepSeek R1 is also available as an open-source model.
Baidu's comeback in the AI competition
Challenging market position
Baidu was the first Chinese technology company to launch a large language model (LLM) in March 2023. Despite this early start, Baidu has fallen behind competitors such as Alibaba, ByteDance, and especially the surprise success of DeepSeek.
The user numbers speak for themselves: ByteDance's chatbot Doubao has the most monthly active users in China with 78.6 million, followed by DeepSeek with 33.7 million, while Baidu's Ernie Bot lags far behind with 13 million.
Strategic importance and global impact
The launch of the new models is part of a catch-up strategy. With its aggressive pricing strategy and planned open-source releases, Baidu could not only regain ground in the Chinese market but also increase the pressure on Western AI labs.
The so-called “deepseek moment”—the demonstration that a Chinese AI lab can achieve similar performance to Western providers with fewer resources—has already caused unease in the industry. This is evidenced by recent letters from the AI labs Anthropic and OpenAI to the US government, in which they call for regulation of AI from China.
Potential and open questions
Baidu's new AI models could significantly alter the dynamics of the global AI market, particularly through the combination of a strong performance promise, aggressive pricing, and planned open-source availability. However, whether Ernie X1 can truly compete with DeepSeek R1, and whether Ernie 4.5 even surpasses GPT-4.5, remains to be seen and must be confirmed by independent testing.
The announcement highlights the increasing competition between Chinese and Western AI providers and could lead to wider availability and lower prices for advanced AI technologies. At the same time, Baidu's open-source approach could fundamentally change the global AI landscape and further accelerate the pace of innovation.
For users and businesses, the new models potentially offer a cost-effective alternative to established services, although their actual performance in practical use remains to be seen. The coming months will show whether Baidu can truly replicate the success of DeepSeek with its new technology and regain lost market share.
Suitable for:
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your national language!
Konrad Wolfenstein
I would be happy to serve you and my team as a personal advisor.
You can contact me by filling out the contact form or simply call me on +49 7348 4088 965 (Munich) . My email address is: wolfenstein ∂ xpert.digital
I'm looking forward to our joint project.
