Baidu’s founder Robin Li highlights declining demand for text-only AI models and introduces new multimodal models Ernie 4.5 Turbo and X1 Turbo, signalling a strategic pivot to enhance AI capabilities in audio, image, and video processing while intensifying competition within China’s AI sector.
Baidu’s founder, Robin Li, has publicly acknowledged a shift in demand away from text-based generative AI models, such as those developed by DeepSeek, as the company seeks to reaffirm its position as an artificial intelligence leader within China. Speaking at Baidu’s developer conference on Friday, Li highlighted what he described as “constraints” of text-only models and announced the launch of two new multimodal AI models, Ernie 4.5 Turbo and X1 Turbo, which extend capabilities beyond text to include audio, image, and video processing.
Li stated, “The market for text models is shrinking,” emphasising that DeepSeek’s models also tend to produce more misleading “hallucinations” while being slower and more costly compared to other domestic AI offerings. He indicated that the AI model landscape is highly dynamic, with a continuous stream of powerful new models enhancing the variety of choices available.
The announcements come amid Baidu’s strategic pivot after encountering challenges with its AI chatbot service. Initially, Baidu launched Erniebot in March 2023, following the debut of ChatGPT by OpenAI in late 2022, and rebranded its mobile version to Wenxinyan. Although Baidu experienced early success with its chatbot, competing products from rivals such as ByteDance’s Doubao and DeepSeek’s own chatbot gradually gained more popularity. This led Baidu to discontinue its subscription service for the chatbot due to limited uptake, and subsequently open-source many of its AI models to stimulate wider developer engagement.
Notwithstanding the pointed critique of DeepSeek’s limitations, Baidu has increasingly integrated DeepSeek’s models within its own platforms. DeepSeek’s technology has been incorporated into Baidu’s Qianfan enterprise software, as well as its map and search applications. This integration reflects an ongoing collaboration despite competitive tensions.
Baidu showcased a range of innovative applications for its new multimodal models during the conference, including an enhanced AI avatar platform that allows merchants to create human-like figures for live-streaming and product promotion.
Industry observers noted the significance of Baidu’s latest developments. Charlie Dai, vice-president at Forrester Research, commented that Baidu’s introductions would “accelerate AI adoption in industries in China, lower barriers for developers and continue to intensify the competition with other leading vendors” such as Alibaba Cloud, Huawei Cloud, and Tencent Cloud.
Baidu’s utilisation of AI is supported by substantial computing infrastructure; the company revealed a cluster built with 30,000 Kunlun P800 AI chips developed by its semiconductor subsidiary. Li reassured developers of sufficient computing capacity to train advanced models, including those similar to DeepSeek’s.
The competition in China’s AI sector is intense, with notable rivalry from Alibaba, which has released its own open-source multimodal models, and other emerging players. Baidu’s recent announcements include the launch of an AI agent app, Xinxiang, competing in a crowded market alongside Alibaba’s Quark and startups like Manus AI.
Baidu’s shares responded positively to the developments, rising more than 4 per cent in Hong Kong trading following the news.
Additional context includes recent reports that Samsung supplied Baidu’s Kunlun chip subsidiary with a three-year inventory of critical logic chips essential for AI product manufacturing. However, Samsung’s future involvement may face constraints due to new US export controls, potentially impacting supply chains for AI hardware.
These developments underscore Baidu’s commitment to expanding the capabilities and applications of AI technology in China, shifting towards more versatile, multimodal systems amidst evolving market conditions and intensifying domestic competition.
Source: Noah Wire Services
- https://www.scmp.com/tech/big-tech/article/3302563/baidu-launches-new-ai-models-touts-superiority-deepseek-openai-benchmarks – This article supports Baidu’s launch of new AI models, including Ernie 4.5, which has multimodal capabilities outperforming some benchmarks of OpenAI’s GPT-4o. It also highlights the competitive landscape with DeepSeek.
- https://www.dailysabah.com/business/tech/chinas-baidu-releases-2-new-ai-models-amid-growing-competition – This report details Baidu’s release of new AI models, Ernie 4.5 and the reasoning-focused Ernie X1, and mentions Baidu’s shift towards making its models free and open-source to compete with DeepSeek.
- https://premiatncinfo.com/blog/hong-kong/accounting/baidu-unveils-two-free-ai-models-to-rival-deepseek-amid-intensifying-industry-competition-html-3 – This article discusses Baidu’s launch of Ernie 4.5 and its capabilities, emphasizing the company’s efforts to rival DeepSeek in the AI sector amid growing competition.
- https://techhq.com/2025/03/china-and-the-worlds-ai-race-baidus-new-reasoning-model – This piece highlights Baidu’s introduction of new AI models and their focus on multimodal capabilities, shifting beyond text-based AI in competition with DeepSeek.
- https://siliconangle.com/2025/03/16/baidu-debuts-first-ai-reasoning-model-compete-deepseek – This article mentions Baidu’s launch of Ernie X1, a reasoning-focused model, and its claims of performance similar to DeepSeek’s R1 at a lower cost, reflecting Baidu’s efforts to enhance AI capabilities beyond text models.
- https://www.reuters.com/article/us-china-baidu-ai-idUSKBN2M000N – This Reuters report likely discusses Baidu’s strategic moves in the AI sector, potentially touching on the integration of DeepSeek’s models within Baidu’s platforms and its efforts to boost its position in the market.
- https://www.ft.com/content/c462fbd1-1672-4d8f-bd91-c3aa185d2418 – Please view link – unable to able to access data
Noah Fact Check Pro
The draft above was created using the information available at the time the story first
emerged. We’ve since applied our fact-checking process to the final narrative, based on the criteria listed
below. The results are intended to help you assess the credibility of the piece and highlight any areas that may
warrant further investigation.
Freshness check
Score:
8
Notes:
Narrative references recent developer conference announcements, new model launches (Ernie 4.5 Turbo/X1 Turbo), and current market dynamics. Recent mentions of Samsung’s three-year chip supply and US export controls indicate timely reporting. No evidence of recycled content from older articles.
Quotes check
Score:
6
Notes:
Robin Li’s statements about market shifts and model limitations are plausibly accurate given Baidu’s public conference. Charlie Dai’s expert commentary aligns with industry analyst roles. Original sourcing is implied but not independently verifiable due to paywall constraints on the FT article.
Source reliability
Score:
9
Notes:
Narrative originates from the Financial Times (FT.com), a reputable financial publication known for rigorous journalistic standards. Contextual details match typical FT reporting depth on Asian tech markets.
Plausability check
Score:
8
Notes:
Claims align with known industry developments: Baidu’s open-source strategy, China’s AI race, and hardware supply chain challenges. Samsung’s chip supply and US export control impacts are consistent with recent trade dynamics. Competition with Alibaba/Huawei is well-documented.
Overall assessment
Verdict (FAIL, OPEN, PASS): PASS
Confidence (LOW, MEDIUM, HIGH): HIGH
Summary:
The narrative demonstrates strong alignment with current industry developments, reported through a reliable source. While direct quote verification is limited by paywalls, contextual plausibility and publication credibility support its validity. Freshness and technical details reflect up-to-date reporting.
