Elon Musk revealed that his xAI platform has open-sourced the Grok 2.5 model, with Grok 3 expected to be released in six months. He also highlighted that Chinese AI companies are his platform’s biggest competitors.
According to xAI, the Grok 2.5 model, which contains 42 files and occupies 500GB, has been made available on Hugging Face for download. xAI’s instructions direct developers to download the files, set up the SGlang inference engine, and launch the inference server with the tokenizer to run the model. The model requires at least eight GPUs, each with more than 40GB of video memory.
Musk confirmed in a post on X that Grok 2.5, which he considers his firm’s best model, has been made available for public use. He added that Grok 3 will also be made available in about six months. He pointed out the competitive nature of the AI landscape by mentioning that Chinese AI companies pose a big challenge for them beyond Google’s competition.
The @xAI Grok 2.5 model, which was our best model last year, is now open source.
Grok 3 will be made open source in about 6 months. https://t.co/TXM0wyJKOh
— Elon Musk (@elonmusk) August 23, 2025
Elon’s post specified that the model requires a significant amount of resources to run. Grok 2.5, which occupies 500GB, will require at least 8 GPUS, each with 40 GB of video memory. Users can access the model weight, which has been made available on Hugging Face for download.
Some developers online have pointed out that the model requirements are too high for most of them to run.
xAI has also released a new update for the Grok app, the latest version being v1.1.58. The update adds AI-powered video generation features, which can now be accessed directly through the platform. Musk confirmed that the company will continue to improve the Grok app and pursue the open-source roadmap simultaneously.
Grok 2, which primarily shaped Grok 2.5, performance data shared by xAI showed Grok 2 achieving results that compete with leading models such as Claude and GPT-4.
On the LMSYS leaderboard, Grok 2’s Elo score exceeded those of Claude and GPT-4. Such benchmarks placed the model in a strong position against postgraduate scientific knowledge (GPQA), general knowledge (MMLU, MMLU-Pro), and mathematical models.
The open-source release has sparked debates online, focusing on the nature of the license and xAI’s failure to disclose the exact parameters of the model on Hugging Face. AI engineer Tim Kellogg described the license as custom with some anti-competitive terms. Companies such as DeepSeek, Qwen, OpenAI, and Microsoft have used more open licenses such as Apache 2.0 and MIT.
The Grok app has faced some controversies in the past, with the most recent being the responses related to the white genocide conspiracy in South Africa. Cryptopolitan reported on it in May, highlighting the incident in which Grok gave responses that question the Holocaust’s death toll and even described itself as “MechaHitler.”
Musk later described Grok 4 as a truth-seeking AI model, although it still faces some controversies. Some users online identified that it referred to Musk’s own social posts when responding to contentious topics.
Chinese tech giants Baidu, Alibaba, and Tencent have also released more than ten model updates since January this year. Baidu expanded its input limit, allowing over 1,000 characters and interacting conversationally with its AI chatbot in the latest update release. Nearly all of their models are open-source, showing China’s ambition to be a global leader in AI. Such firms pose a great challenge for Western tech firms like OpenAI, Google, and xAI.
Want your project in front of crypto’s top minds? Feature it in our next industry report, where data meets impact.