Baidu’s ERNIE launches ‘signal a global AI race’

The launch this weekend by Baidu of a native open-source multimodal foundation model called ERNIE 4.5 and a reasoning model, ERNIE X1, could lower AI adoption barriers, intensify competition, and reshape pricing strategies across the industry, an analyst said Monday.

In order for that to occur, said Thomas Randall, research lead for AI at Info-Tech Research Group, “the success of these models will depend on performance validation, developer adoption, and enterprise trust. However, they signal a global AI race where cost-efficiency and accessibility become as important as raw capability.”

The Chinese tech giant said in a release that the introduction of the two offerings “pushes the boundaries of multimodal and reasoning models,” adding that ERNIE X1 “delivers performance on par with DeepSeek R1 at only half the price.”

Baidu said it plans to integrate both new ERNIE (Enhanced Representations through Knowledge Integration) models into its product ecosystem, and that the integration will include Baidu Search as well as other offerings.

In addition to the reasoning model, Baidu said that its ERNIE Bot large language model (LLM), is “being made free to the public ahead of schedule.” Its original plan was to make ERNIE Bot fully free to individual users starting April 1.

For enterprise users and developers, it said, ERNIE 4.5 is now available via APIs on Baidu AI Cloud’s MaaS platform, Qianfan, while ERNIE X1 is scheduled to be available on the platform “soon”.

Jason Andersen, an analyst with Moor Insights & Strategy, said, “while the price-to-performance of training the model is a consideration, it’s unclear if that is enough to differentiate for a developer or an enterprise.”

For example, he said, “this model, like DeepSeek, is open source. Therefore, we will likely see AI hosting providers pick it up alongside other models. Once it’s in the AWS or Azure toolsets, will it live up to the promise?”

“To be more specific,” he added, “which tasks will it be best suited for? Also, there will be the customer’s ability to work with the model and get assistance as it gains adoption. I am of the opinion that we have moved beyond the benchmark drag race and need to consider business sustainability and value.”

According to Randall, Baidu’s plan to open-source ERNIE 4.5 “may accelerate innovation even further via community-driven improvements, challenging the closed-source dominance of OpenAI and Google. Indeed, with China’s regulatory push for domestic AI dominance, Baidu’s progress strengthens the nation’s AI independence from US-based models.”

He said, “regional competition may intensify, with Alibaba, Tencent, and ByteDance almost certainly looking to respond by investing further in this technology. Western firms may face new restrictions or competition in China, as businesses opt for homegrown models for compliance reasons.”

Baidu said in the release that ERNIE 4.5 achieves the following via what it refers to as refined language skills:

Improvements in “understanding, generation, reasoning and memory, along with notable enhancements in hallucination prevention, logical reasoning, and coding capabilities.”

The ability to integrate and understand text, images, audio and video content as well as “outperform GPT-4.5 in multiple benchmarks while priced at just 1% of GPT-4.5.”

Reasoning model ERNIE X1, Baidu said, supports “various tools including advanced search, Q&A on given document, image understanding, AI image generation, code interpreting, webpage reading, TreeMind mapping, Baidu academic search, business information search, and franchise information search.”

The release of both comes 10 days after China’s Alibaba Cloud launched QwQ-32B, a compact reasoning model built on its latest large language model (LLM) Qwen2.5-32b, one it says delivers performance “comparable to other large cutting edge models” including those of Chinese rival DeepSeek and OpenAI’s ol  “with only 32 billion parameters.”