The most powerful Grok 3 model that xAI launched a few days ago was initially only available to X Premium+ subscribers, with no free access. We also did a special feature on a way you could try it out. Early-Grok-3 The website for the early model. Perhaps hoping to attract more people to pay for subscriptions, xAI earlier suddenly announced on X that “Grok 3 is free to use,” and even opened up DeepSearch and Think features. Those who want to try it should seize the opportunity.
xAI announced on social media that Grok 3 is free for a limited time, until the servers crash.
xAI announced earlier on X: “The world’s smartest AI: Grok 3 is now free to use (until our servers can’t take it anymore).”
This is it: The world’s smartest AI, Grok 3, now available for free (until our servers melt).
Try Grok 3 now: https://t.co/Tj0afLoxEz
X Premium+ and SuperGrok users will have increased access to Grok 3, in addition to early access to advanced features like Voice Mode pic.twitter.com/YgKavSCiWr
— xAI (@xai) February 20, 2025
From this description, it seems like a limited-time release. However, when I used it, I noticed that the Grok 3 Beta model was the one available. Also, the latter part of the announcement mentioned: “X Premium+ and SuperGrok users will gain access to a more advanced Grok 3 and can try out advanced features early, such as voice mode.”
Therefore, I’m guessing that maybe the free version will just stay on Grok 3 Beta indefinitely. Plus, OpenAI has also teased that the GPT-5 coming out in a few months will be available on the free tier too, so to compete with OpenAI, xAI probably wouldn’t leave free users stuck on the previous generation model.
Grok 3 Beta is now available on X, the web version, and the app. If you don’t see it in the app, please update to the latest version.
Here I am X social platformAs a demonstration, after entering the Grok chat interface, you’ll see the model at the top becomes Grok 3 beta, and below the input box there are also newly added DeepSearch and Think features:
You can ask in Chinese, but the default response is in Simplified Chinese. If you want it to reply in Traditional Chinese, start with a sentence like “用繁體中文回答” at the beginning.
DeepSearch actually works in real testing — it performs multi-step deep searches, browses a large number of web pages, and then provides an answer. For example, when I asked “how to improve work efficiency,” it searched up to 79 web pages.
Some key citations will be added at the end of the response for our reference and verification.
The Think feature makes Grok 3 think a bit longer, doesn’t rush to give an answer, and also shows its reasoning process.
Grok 3 beta also supports file uploads:
Of course, it also supports a search function, but I don’t think it’s very good — ChatGPT, Perplexity, and Felo are better:
As for the other newly added Big Brain feature, it requires the full Grok 3 model, so it has not been made available to the Grok 3 Beta model.
The following chart is the test data shared at the previous launch event. Grok-3 led its competitors in all tests, including GPT-4o, Claude 3.5 Sonnet, DeepSeek-V3, and Gemini-2 Pro:
Reasoning and test-time compute likewise surpass the o3mini (high) and o1 models:
Source: KOCPC Chinese