BEIJING, Feb 27 (Reuters) – Chinese tech giant Baidu (9888.HK) is preparing to release the next iteration of its artificial intelligence model, Ernie, in mid-March, according to a source with direct knowledge of the matter.
The upgraded Ernie 4.5 model is expected to feature significant improvements in its reasoning capabilities. Furthermore, the model will offer enhanced multimodal functions, allowing it to process and integrate various data types, including text, video, images, and audio.
Baidu had previously stated that the Ernie 4.5 series would be gradually rolled out in the coming months, with an official open-source release scheduled for June 30. Baidu has not yet issued an official comment on the upcoming launch.
This move follows increasing competition from other Chinese AI developers, particularly DeepSeek, which has been aggressively releasing AI models that claim to rival or exceed the performance of leading US models at a lower cost. The emergence of DeepSeek has intensified the global AI race and presented a challenge to Baidu’s position.
Despite being one of the first Chinese tech companies to launch a ChatGPT-style chatbot, Baidu’s Ernie large language model has faced challenges in achieving widespread adoption, despite the company’s claims of performance comparable to OpenAI’s GPT-4. The rise of DeepSeek has added further pressure on the Beijing-based company’s AI ambitions.
Baidu CEO Robin Li previously advocated for keeping AI models closed-source, believing it was the best approach for development within the sector. However, DeepSeek’s success has prompted a shift in strategy. During an earnings call earlier this month, Li described Ernie 4.5 as the company’s “best model ever,” acknowledging that developments in the market had shifted the company towards an open-source approach.
In a related development, Alibaba recently announced its intention to open-source its video and image-generating AI model, Wan 2.1, further intensifying the competition in the sector.
