Elon Musk’s xAI is generating buzz tonight with the imminent release of their upgraded large language model, Grok 3.5. Last week Musk suggested Grok 3.5 would be released to early access users this week.
Grok 3.5 has emerged first as a new account on X. @Gork (as opposed to @Grok) currently has 80k followers at the time of writing and growing fast after Elon invited people to try Gork.
While not yet available as a model selection at grok.com or via X.com, you can get an early look at the capabilities by messaging or replying to the @Gork account.
Firstly, you should know that this account is listed as a ‘Parody account’ and based on the replies, it’s clear that Grok 3.5 model is fixed in ‘Unhinged Mode’ as the replies are cheeky at best and blunt, rude or offensive at worst. It is worth remembering this before making an assessment of it’s capabilities as the full release will have different modes, just as Grok 3 has now.
This direct interaction on social media provides a unique way for potential users to get a feel for Grok 3.5 before its wider release.
I did ask Gork to write me a 1,000 word reply to which it refused.
I also tried replying to Gork for a while to see if there was any kind of limit before it got bored and it does seem that after about 5 replies, it went quiet.
It looks like Gork (aka Grok 3.5) has the ability to reply in multiple languages.
As I continued to try the model, I started to see delays in the replies, so I asked about it.. then got this reply.
First Principles Reasoning
Grok 3.5 is touted to excel in “first principles” reasoning. This means it aims to derive answers to complex technical queries, such as those related to rocket engine cycles or electrochemistry, from foundational knowledge rather than simply retrieving information from the web.
Availability for SuperGrok Subscribers
The initial beta release of Grok 3.5 is expected to be rolled out to xAI’s top-tier SuperGrok subscribers. This exclusive early access will allow these users to experience the advancements firsthand.
Challenging the Competition
The arrival of Grok 3.5 puts further pressure on other major players in the LLM space, such as OpenAI, Anthropic, and Google. The ability to reason from first principles could be a key differentiator, potentially allowing Grok 3.5 to tackle novel and highly technical questions with greater accuracy.
The official launch of Grok 3.5 will be closely watched by the tech community as the performance of Grok 3 seen it shoot to the top of the LLM leaderboards in real-world scenarios and against established benchmarks. With more compute and more training data, Grok 3.5 is expected to be another big leap in what’s possible with large language models.
Update: It seems some lucky users did get access to Grok 3.5.
For more information, head to https://x.ai/grok
