US tech billionaire Elon Musk’s artificial intelligence company, xAI, has released Grok 2.5 as open source, a move that intensifies the global race between Washington and Beijing over control of the future of artificial intelligence.
Musk announced over the weekend on X.com that Grok 2.5—described as xAI’s “best model last year”—has been uploaded to Hugging Face, the world’s largest platform for open-source AI. He also disclosed that Grok 3, xAI’s current runner-up to its most advanced Grok 4, would follow within six months.
The timing of the release underscores the rising competition between US and Chinese AI developers. Just weeks earlier, OpenAI launched two open-source models—GPT-OSS-120b and GPT-OSS-20b—in what industry analysts described as a direct challenge to Chinese dominance in the open-source AI ecosystem. Alibaba and DeepSeek, both Chinese firms, have so far led the field with freely accessible AI models that are widely adopted by enterprises and research institutions worldwide.
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).
Zhou Hongyi, founder and CEO of Chinese cybersecurity company Qihoo 360, said on Sunday, according to SCMP, that China’s early embrace of open-source AI has forced US labs to adjust their strategies.
“Open-source is not a trade-off; it’s a more powerful strategy that will help open models outcompete closed ones in the long run,” Zhou said in a social media post, adding that platforms such as DeepSeek have attracted not just Chinese but also American companies and universities.
xAI’s move also reflects how quickly AI models can lose their cutting edge. When Grok 2 launched in August last year, its reasoning and coding abilities ranked it the world’s second-best AI model, only behind OpenAI’s GPT-4, and tied with Google’s Gemini 1.5 Pro, according to benchmarking site LMArena, developed by University of California, Berkeley scientists. But in less than a year, Grok 2 slipped dramatically, now sitting at 59th place. Grok 2.5 fares only slightly better.
In contrast, Grok 3 currently holds 13th position, while xAI’s flagship Grok 4, released last month, is still closed-source but freely accessible to users via the Grok chatbot.
Some observers, however, questioned whether Musk’s open-sourcing of an outdated model carries real weight. Grok 2.5 has been released with restrictions that prevent developers from training new models or building derivative systems on top of it. An earlier iteration of the license even limited commercial use to companies earning under US$1 million annually, requiring others to apply for a separate license.
This guarded approach sets xAI apart from Chinese rivals, which often use permissive licenses such as Apache 2.0 that grant global developers the right to freely use, modify, and commercialize their models. By comparison, critics argue that xAI’s licensing undermines the very principle of open source, even if it represents a symbolic gesture in the geopolitical race.
The broader contest has become about more than just technology. China’s aggressive push to democratize AI by making its models freely available has given it an edge in adoption and global influence, while US firms—often reliant on proprietary strategies—are being nudged into experimenting with open access to avoid ceding ground.
With Musk promising to open Grok 3 in the coming months, the spotlight now shifts to how far xAI and other US players are willing to go in balancing commercial protection with the global momentum toward fully open AI ecosystems.
OpenAI CEO Sam Altman has acknowledged that China’s surge in open-source artificial intelligence played a key role in the company’s decision to release its own open-weight models earlier this month.
Speaking during a media briefing, Altman said the rise of Chinese-developed models was a decisive factor. The launched of the two open-weight models—gpt-oss-120b and gpt-oss-20b—marking its first return to publicly available model weights since GPT-2 in 2019, on August 5, enables developers to download, run, and fine-tune the models locally, a sharp contrast to the company’s closed releases of GPT-3, GPT-4, and the current GPT-5, which remain proprietary.
Altman said on X that the models were meant to align with OpenAI’s mission “to ensure AGI that benefits all of humanity.”
“It was clear that if we didn’t do it, the world was gonna be mostly built on Chinese open-source models. That was a factor in our decision, for sure. Wasn’t the only one, but that loomed large,” CNBC quoted him as saying.