Baidu to open-source its Ernie LLM series
Baidu just announced plans to open-source its upcoming Ernie 4.5 large language model series on June 30. The models will roll out to users over the coming months before the code is released. The shift comes shortly after DeepSeek-R1 shook up the industry, putting pressure on tech giants to rethink their open-source AI strategies.

For the past two years, Chinese tech giant Baidu has firmly defended the closed-source camp, betting its future on proprietary models and its Ernie Bot chatbot ecosystem.
Now, the playbook has officially changed.
Baidu announced plans to open-source its Ernie family of large language models later this year. According to Reuters, the move focuses on the upcoming Ernie 4.5 series, which will roll out to users in the coming months before its code is released to the public on June 30.
The roadmap from Ernie 3.0 to Ernie 5
Ernie powers Baidu's ChatGPT competitor, Ernie Bot, which reportedly hit 430 million users as of November, up sharply from over 200 million in April.
The service first launched in March 2023 on top of Ernie 3.0, a 10-billion-parameter model trained on a 4-terabyte dataset that Baidu began building back in 2019.
Its current flagship, Ernie 4.0, expanded into multimodal capabilities across text, image, and video generation, along with specialized reasoning improvements like solving geometry problems.
Ernie 4.5 is slated to be followed by an even more powerful generation, Ernie 5, targeted for release in the second half of 2025. Reuters notes it is still unclear what capabilities Ernie 5 will deliver and whether it will also be open-sourced.
The DeepSeek ripple effect
Baidu's strategic pivot follows hot on the heels of DeepSeek-R1, the Chinese open-source reasoning model that shocked the AI world weeks ago by matching or beating OpenAI's o1 on key benchmarks and triggering a short-lived Nvidia selloff.
The shockwaves are hitting Silicon Valley as well. In a Reddit thread on Jan. 31, OpenAI CEO Sam Altman admitted the company will need to figure out a different open-source strategy, though he noted internal dissent and said it is not currently OpenAI's highest priority.
Why it matters
Baidu going open-source is a clear signal that closed-wall dominance is losing its defensive moat.
For enterprise developers, having an open-source model from China's primary search player could provide a robust alternative for Chinese-language workflows, provided licensing and resource requirements are practical. If you run heavily customized enterprise stacks, it is worth watching.
However, we think Baidu's timeline carries real risk. June 30 is practically an eternity in current AI cycles. By mid-year, the open-weight bar will have moved again, and releasing yesterday's weights could feel anticlimactic.
It also reveals an awkward hedging strategy. Open-sourcing Ernie 4.5 while holding back Ernie 5 shows Baidu is still reluctant to go all-in on open source. If Ernie 4.5 does not deliver genuine state-of-the-art efficiency, Baidu risks giving away IP without earning the community developer goodwill it desperately needs.
Sources (1)
- siliconangle.com(opens in a new tab)Originalsiliconangle.com