As the rivalry for large language models (LLMs) heats up, Chinese tech giant Baidu unveiled two new AI models that it claimed were superior to DeepSeek and OpenAI based on certain criteria.
As it tries to differentiate itself in a very competitive AI race, China’s Baidu, which opens a new tab, said on Sunday that it has introduced two new AI models, including a new reasoning-focused model that it claims rivalled DeepSeek’s model.
The industry has been agitated and the global AI race has been reignited by the introduction of AI models by Chinese AI firm DeepSeek, which claims that its models are on par with or even better than industry-leading models in the US at a fraction of the cost.
On Sunday, Baidu released Ernie X1, its first multimodal reasoning model, and Ernie 4.5, its most recent multimodal fundamental model, for free on its website.
With regard to one of the new models, “ERNIE X1 delivers performance on par with DeepSeek R1 at only half the price,” Baidu stated in reference to one of the new models introduced. Baidu said that the X1 is the first deep thinking model to employ tools on its own and that it has “stronger understanding, planning, reflection, and evolution capabilities.”
According to Baidu, ERNIE 4.5, its most recent foundation model, has “excellent multimodal understanding ability.” Its linguistic skills are more sophisticated, and its comprehension, creation, reasoning, and memory have all been much enhanced.
In a statement shared on WeChat, Baidu claimed that Ernie 4.5’s multimodal capabilities—which include picture, audio, and video—surpassed OpenAI’s GPT-4o on a number of benchmark platforms, including CCBench and OCRBench. According to the business, the text skills of the Ernie 4.5 basic model also outperformed DeepSeek V3 on a number of benchmarks and were comparable to the GPT-4.5 from US startup OpenAI.
Additionally, Baidu released the models for free more than two weeks early. Previously, access to the company’s most recent AI models required a monthly fee.
After the flurry caused by the introduction of OpenAI’s ChatGPT, Baidu became the first Chinese Big Tech company to introduce an LLM in China in March 2023. However, in the past two years, China’s other AI firms have challenged its early-mover advantage. DeepSeek has sparked an open-source tsunami, with Alibaba, Tencent, and Bytedance sweeping the market for corporate and consumer customers for their models. This is the search giant’s most recent attempt to become more relevant in China’s AI sector.
Although the Beijing-based startup was among the first in China to officially launch a generative AI platform in 2023, competitor chatbots from businesses like Moonshot AI and ByteDance, the owner of TikTok, have since attracted more users.
According to Baidu, it also has “high EQ” and is simple to comprehend sarcastic cartoons and network memes.
In the consumer-facing AI space, where Baidu faced fierce competition, the company DeepSeek upended the market both domestically and internationally with a model that outperformed rivals like ChatGPT, manufactured in the US, but was far less expensive to produce.
Baidu claimed that their new reasoning model, Ernie X1, “delivers performance on par with DeepSeek R1 at only half the price,” however it did not disclose benchmark data.
Despite boasting performance equivalent to OpenAI’s GPT-4, Baidu, one of the first tech firms in China to introduce a chatbot like ChatGPT, has had difficulty gaining traction for its Ernie big language model in the face of intense competition.
While other software firms have been catching up, Chinese businesses and local government organizations have since hurried to integrate DeepSeek’s open-source methodology into their projects.
The R1 reasoning model from DeepSeek has been included into Baidu’s search engine.
Tencent, the company that owns WeChat, introduced a new AI model in February that it said could answer questions more quickly than DeepSeek, despite integrating its competitor’s technology into its messaging app.
Text, audio, video, and picture data may all be processed and integrated by multimodal AI systems, which can also translate material between different formats.
According to the startup, commercial users can pay 2 yuan (US$0.28) per million token inputs and 8 yuan per million token outputs for access to Ernie X1’s application programming interface (API).
DeepSeek-reasoner, which is driven by its reasoning model R1, is now priced at US$2.19 per million token outputs and US$0.55 per million token inputs. Due to a spike in demand, the Hangzhou-based startup increased the price of its API last month.
In a complete reversal from his long-standing support for closed-source AI development, Baidu founder, chairman, and CEO Robin Li Yanhong announced last month that Ernie 4.5 will be open source starting on June 30.
During a February earnings call with investors, Li stated, “One thing we learned from DeepSeek is that open sourcing the best models can greatly help adoption.” “People are naturally curious and want to try the model when it is open source, which helps drive broader adoption.”
Beijing-based Baidu’s company is suffering from low ad income despite its advancements in AI. According to its announcement last month, total revenue for the fourth quarter decreased by 2% year over year, and full-year revenue decreased by 1%.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
