Baidu is trying to make price, access, and open-source availability central to its next phase in artificial intelligence. The Chinese tech company has introduced Ernie X1 and Ernie 4.5, two models aimed at a market where performance claims are now being weighed against API cost and distribution strategy.
The headline claim is direct: Baidu says Ernie X1 performs on par with Deepseek-R1 while costing 50 percent as much. The company has not yet released benchmarks or technical details to support that comparison, so the claim remains one to watch rather than a settled finding.
What Baidu Is Claiming About Ernie X1
Ernie X1 is Baidu's new reasoning model. The company is positioning it against Deepseek-R1, the model that has become an important reference point for lower-cost, high-capability AI systems.
According to Baidu, Ernie X1 costs $0.28 per million input tokens and $1.10 for output through its Qianfan cloud platform. The source article states that this is half the API cost of Deepseek-R1, while also noting that Deepseek-R1 is available as open source.
That distinction matters. A lower hosted API price can appeal to companies that want managed access. Open-source availability can appeal to developers and organizations that want more control over deployment. Baidu has not announced open-source plans for X1, but the source article says such a move appears likely because Deepseek-R1 is already open source and Baidu currently uses Deepseek-R1 in some of its products.
For now, the key limitation is evidence. Baidu has made a performance and pricing claim, but without published benchmarks or technical details, outside users cannot fully evaluate how Ernie X1 compares with Deepseek-R1 across real workloads.
Ernie 4.5 Expands Baidu's Multimodal Push
Baidu also unveiled Ernie 4.5, a multimodal base model designed to process videos, photos, and text more effectively than earlier versions. The company says the model improves on hallucination reduction, complex reasoning, and code writing.
Internal benchmarks from Baidu suggest Ernie 4.5 performs better than OpenAI's standard GPT-4o and matches GPT-4.5. As with Ernie X1, these are company claims from internal benchmarks, so the public picture depends on what Baidu releases and how the model performs in external use.
The pricing comparison is more concrete. Through Qianfan, Baidu lists Ernie 4.5 at $0.55 per million input tokens and $2.20 for output. The source article compares that with GPT-4.5's rates of $75 per million input tokens and $150 per million output tokens.
That gap is the central business signal. Baidu is not only introducing another model family; it is presenting model access as a cost challenge to Western AI companies with higher pricing for comparable capabilities.
Open Source Marks A Shift In Strategy
Baidu's approach is also changing on openness. Mid-February, Baidu CEO Robin Li announced that the company will make its Ernie 4.5 series open-source starting June 30.
That is notable because Li had historically supported keeping models closed. His stated reason for the shift was that open source would "spread the technology much faster." In practical terms, the move puts Baidu closer to the strategy that helped Deepseek-R1 draw attention beyond China.
The source article frames this as part of the broader pressure created by the "Deepseek moment." In January 2025, Deepseek demonstrated that it could match the performance of Western AI labs with significantly fewer resources. Since then, Chinese open-source models have become a commercial concern for companies that sell access to closed models through APIs.
Open source can change how quickly developers test, adapt, and compare AI systems. It can also make pricing harder to defend when alternative models offer similar capabilities at lower cost or with more flexible access.
Why This Pressures Western AI Companies
Baidu's new releases land in a market already sensitive to cost. The source article says Western AI companies have taken notice, with Anthropic and OpenAI petitioning the US government to regulate Chinese AI development because of national security and political concerns.
The commercial issue is also clear from the source: open-source Chinese models could significantly affect these companies' API businesses. If users can access lower-priced models or open-source alternatives that meet their needs, higher-priced hosted services face a harder competitive argument.
Baidu is not new to this race. As China's dominant search provider, it moved early with Ernie Bot, described in the source as China's first ChatGPT alternative. It launched in 2023 and drew 70 million users within three months.
The company is now making Ernie Bot available at no cost to individual users. It also plans to bring both Ernie X1 and Ernie 4.5 into existing products, including Baidu Search and the Wenxiaoyan app.
Taken together, these moves show a company trying to compete on several fronts at once: reasoning performance, multimodal capability, API pricing, product integration, and open-source distribution. The unresolved question is how Baidu's performance claims hold up once the models are tested outside internal benchmarks.
The Bottom Line
Baidu's Ernie X1 claim is bold but still incomplete without public benchmarks or technical details. Its pricing, however, is already part of the story: $0.28 per million input tokens and $1.10 for output places it directly against Deepseek-R1 on cost.
Ernie 4.5 adds another layer. Baidu is presenting it as a stronger multimodal model, pricing it far below GPT-4.5 on Qianfan, and preparing to open-source the series starting June 30.
If Baidu follows through and outside testing supports its claims, these releases could intensify the pressure already created by Deepseek. For users and businesses, the immediate question is practical: whether lower-cost and open-source Chinese AI models can deliver enough performance to change buying decisions.