Baidus Shift: Unveiling the Open Source Code of Ernie AI Model

The Chinese tech giant Baidu has announced plans to open the source code of its generative AI model, Ernie. Previously, the company had been opposed to open-source policies.

«Baidu has consistently upheld its proprietary business model and resisted open-source initiatives, yet rule-breakers like DeepSeek have demonstrated that open-source models can be as competitive and reliable as proprietary ones,» stated Lian Jie Su, chief analyst at research and consulting group Omdia.

Researchers consider Baidu’s decision a pivotal moment in the global AI race.

«Whenever a major lab releases the source code of a powerful model, it raises the standards for the entire industry,» said Sean Ren, an assistant professor of computer science at the University of Southern California and AI Researcher of the Year at Samsung. He believes that Baidu’s actions will pressure supporters of closed models like OpenAI and Anthropic.

«While most consumers may not care whether a model’s code is open or not, they are concerned about lower costs, better performance, and support for their language or region. These advantages often come from open models that allow developers and researchers greater freedom for iteration, customization, and faster deployment,» Ren explained.

Other industry experts view the release of Ernie’s code as a challenge to competitors in the U.S. and China regarding pricing strategies. «Baidu has essentially thrown a ‘Molotov cocktail’ into the AI world. OpenAI, Anthropic, DeepSeek, all those players who thought they were selling top-notch champagne are about to realize that Baidu is willing to offer something equally powerful,» remarked Alex Strathmore, founder of consulting firm Epic Loot. He added that «this isn’t a competition; it’s a declaration of war on pricing.»

In March, Baidu announced that the Ernie X1 model delivers performance comparable to DeepSeek R1 at «just half the price.»

Baidu’s CEO Robin Li hinted earlier this year that the rollout of Ernie will assist developers globally in AI development. «Our releases are aimed at empowering developers to create superior applications without worrying about model capabilities, costs, or development tools,» he explained.

Cliff Yurkevich, Vice President of Global Strategy at applied AI firm Phenom, compared the shift to an open architecture in AI to the difference between Android and Apple. «When Android first launched, its hallmark was that it could be customized and tailored. However, this required significant effort; people just wanted the OS to function properly,» the expert stated.

OpenAI’s CEO Sam Altman previously indicated that the organization might shift towards open-source deployment this summer. «Although OpenAI had open-source models in the past, the company has generally favored a proprietary development approach. I believe we need to devise a different strategy around open source. Not everyone at OpenAI shares this view, and it isn’t our current top priority… We’ll develop better models in the future, but we’ll have less dominance than we did in previous years,» he noted.

Recently, Altman announced that the release would be postponed but is still scheduled for summer.

According to Yurkevich, one advantage of U.S. players is that Baidu’s model may face skepticism at the enterprise level and numerous security concerns.

Ren affirms these worries: «Just because the model weights are public doesn’t mean we know what data it was trained on, whether consent was obtained, or if compensation was paid to the creators.» He added, «If we don’t address this issue now, we risk scaling systems that silently extract value from the data of millions without their consent or compensation.»

Strathmore also remarked that as products become tied to Baidu’s APIs, «China will effectively gain access to every application on every phone.»