Hey everybody, welcome back to the Elon Musk Podcast. This is a show where we discuss the critical crossroads, The Shape, SpaceX, Tesla X, The Boring Company, and Neurolink. I'm your host Will Walden, Elon Musk's AI company XAI has released Grok 2.5 Urban Open Source License, raising one question what exactly does Elon Musk gain from giving away his
code? This makes XAI the first major U.S. company to open source, a flagship conversational model while it's still considering competitive and it directly challenges the closed source strategies of other companies like Open AI, Anthropic and Google DeepMind. Musk framed the decision as a philosophical one about transparency, but the decision also positions XAI to expand its influence without outspending its rivals and user acquisition
or cloud infrastructure. Now. XAI published the weights and architecture for Grok 2.5 Friday night, uploading them to a public GitHub repository under the Apache 2 Point O license. A license allows anyone to use, modify, or build on Grok 2.5, including for commercial purposes, without paying XAI anything or attributing anything to XAI. Now this immediately makes Grok 2.5 one of the most permissively licensed large language models
ever released at this scale. The model comes in two sizes, a 314 billion parameter version trained with mixture of experts techniques and a smaller dense version with 25 billion parameters. Both are trained on publicly available and licensed data sources, according to X AI and Elon Musk described the release as a commitment to openness and said it aligns with XA is broader mission to develop AI
that benefits all of humanity. He previously criticized Sam Altman and Open AI for switching to a closed source model after becoming a for profit company in March. XAI open source Croc 1, though that model lagged way behind current commercial systems, Croc 2.5 narrows that gap
considerably. According to internal evaluations, Croc 2.5 ranks close to GPT 4 on several benchmarks, including MMLU, which is massive multitask language understanding and Human Evil, a programming test commonly used to measure LLM coding capabilities. So releasing Grok 2.5 under Apache 2 Point O immediately places pressure on Sam Altman and Open AI in Anthropic, who have argued that safety risks justify keeping their most
capable models closed. Musk has openly dismissed that rationale, claiming that companies are using safety as an excuse to maintain control over their code base and their user base. Now by Elon open sourcing Grok 2.5 X AI makes it easier for startups, researchers and other companies to train on a strong base model without building from scratch. The trade off Less control for broader adoption mirrors what Meta has done with Llama two and
three. Though Meta still uses a non commercial license, the release also serves A practical purpose. Grok runs inside X, where it appears the chatbot for premium Subs. Making Grok 2.5 open source gives Musk access to a global pool of developers who can test, improve, and adapt the system without being on XA is payroll. Now. That outsourcing effect has already benefited Llama Meta's ecosystem, and Musk appears to be betting that it can be done with Grok as well.
XAI has stated that future versions, including Grok 3, may not be open source, suggesting that Grok 2.5 may serve as the platform's long term public version while development continues behind the scenes on other models. Now, Musk has also said the Grok 2.5 has been trained to be maximally truthful, quality that remains difficult to measure. XAI claims that Grok now outperforms Claude 2.1, Gemini 1.5, and GPT 4 in some conversational tasks, though external benchmarks vary on that.
Grok 2.5 is not connected to Real Time Search or X's live data feed, though that's unlike the private chat bot used by X users. That means developers using the open source Grok will not have access to Musk's internal data pipelines, which remain a competitive asset unique to the version of Grok embedded within X. And open sourcing Grok 2.5 also gives XAI a way to compete with Big Tech without relying on partnerships.
Unlike Anthropic, which has tied itself closely to Amazon, or Open AI, which remains linked to Microsoft, XAI operates as an independent entity that would benefits from its access to X's user data in its upcoming integration with Tesla's interface hardware. Now, the open source model allows XAI to sidestep capital constraints by letting external users do much of the refinement and experimentation that would otherwise require dedicated in house teams.
I want to know if you're on your podcast platform right now. If you have comments, leave a comment. Would you use grok at home? Would you be able to mix it yourself? Would you download it from GitHub and be able to use it? Let me know. Let me know what what comes of this.
So let's get back into Grok. Though the 2.5 version mixture of Experts architecture divides the model into separate subnetworks, only a portion of which activate at any given time, this design improves efficiency and reduces training costs, especially when deployed at scale. It also means that developers who download Grok 2.5 will need infrastructure capable of handling that complexity.
That may limit adoption among individual users or small teams, but larger labs and cloud based EI platforms can run the model with relative ease and for free. Releasing such a large scale open source model at this time sends a huge signal to the industry. XAI is betting on diffusion over exclusivity. By removing licensing friction, it creates a channel for grok based variance to proliferate across research labs, startups and independent projects.
Musk appears to believe that wide availability will help create loyalty and technical entrenchment, even if it's limits of the company ability to monetize Grok directly and make money from every user. The decision also reinforces Musk's public feud with Open AI and Sam Altman. He Co founded the company in 2015, then left in 2018 after
internal disagreements. Since then, Musk has attacked Open AI repeatedly, suing it earlier this year and accusing CEO Sam Altman of betraying its founding principles. XA is release of Croc 2.5 under a fully open license marks a clear ideological split from the commercial closed source path that Open AI has taken since its partnership with Microsoft, and Croc 2.5 is now freely available to download on GitHub.
You can go do it right now. They have install instructions and model weights posted alongside this license. Any developer can fine tune it for their own applications. You can use it for whatever you want to use it for and they have no say over what you do with it. Download the code, you can do whatever you want to with it now. You can integrate it into custom pipelines or study its architecture for academic purposes. You can make a for profit company using Grok.
XAI has not imposed any restrictions on use beyond what is standard under Apache 2 Point O, which includes liability disclaimers. But there's no functional limitations on you. You can just go wild and build whatever you want to build. They're planning a flag right now in a part of the EI space that remains divided between access and control. The release turns Grok into a public tool now that does something you get on X or in your Tesla. It's not just a product anymore.
This is for everybody. And Musk claimed this earlier this year that he was going to release Grok open source and everybody would be able to use it. Now, like we said before, the next version of Grok will be closed source. They're planning on using Grok 2.5 for testing and so they can have something available for people to use and when they upgrade from the next version of Grok to two versions after or even a version after, we might
see that on GitHub as well. I'm an open source proponent. I've used it for thousands of development jobs in the past and still doing that now. So I think this is a great move by X AI. Let me know what you think in the comments and I'll see you in the next one. Hey, thank you so much for listening today. I really do appreciate your support.
If you could take a second and hit the subscribe or the follow button on whatever podcast platform that you're listening on right now, I greatly appreciate it. It helps out the show tremendously and you'll never miss an episode. And each episode is about 10 minutes or less to get you caught up quickly. And please, if you want to support the show even more, go to patreon.com/stagezero and please take care of yourselves and each other and I'll see you tomorrow.