Welcome to Neural. AI moves fast. We help you keep up. OpenAI says GPT-4.5 is coming to ChatGPT in a matter of weeks. But first, xAI will unveil Grok 3 tonight. Now the countdown is on for OpenAI to do the funniest thing ever…
Grok 3 demo day
Fresh off the heels of threatening to buy OpenAI, Elon Musk announced on Saturday that Grok 3 will arrive tonight in the form of a live demo scheduled for 8 p.m. PT.
Musk hypes up Grok 3 by calling it the “smartest AI on Earth.” Actual details around Grok 3 remain a mystery until tonight’s reveal.
Here’s a recap of xAI and Grok’s recent history:
- xAI created in July 2023
- Grok-0 completed in August 2023
- Grok-1 previewed in November 2023
- Grok-1.5 completed in March 2024
- Grok-2 released in August 2024
In 2025, xAI has taken Grok from a feature within the X app to a standalone app and website.
My biggest question about Grok: When will it come to Tesla? AI-driven (supervised) self-driving is one thing, but drivers could benefit from a Grok-powered voice assistant that replaces the basic voice control function.
Question number two is whether or not OpenAI will steal xAI’s thunder tonight and do something fun with GPT-4.5.
This week in OpenAI
OpenAI’s Sam Altman laid out the roadmap for ChatGPT last week. Altman said GPT-4.5 is coming in a matter of weeks, which suggests a release today is unlikely, followed by GPT-5 in a matter of months.
GPT-5 is the more ambitious model that aims to unify fast large language models with slower thinking reasoning models — no need to choose between a half dozen models per query.
Meanwhile, Altman continues to vaguely tease the benefits of GPT-4.5 on X:
When asked to “steal the show” tonight, Altman responded diplomatically with “that wouldn’t be very nice…”
Separately, OpenAI has deployed some actual releases over the last few days:
- OpenAI “o1 and o3-mini now support both file and image uploads in ChatGPT”
- OpenAI “raised o3-mini-high limits by 7x for Plus users to up to 50 per day”
- OpenAI released a Model Spec update that describes how it wants its AI models to behave
Then there was this all-time winner for most vague hype post ever:
The update, apparently, is meant to make GPT-4o a better writer, especially when given examples to follow, and less of an AI-slop generator.
If nothing else, the last week in OpenAI model updates shows just how critical GPT-5 will be for ChatGPT. The flow chart for which models are best at what and how they’re limited is getting beyond unwieldy, and it’s outdated before anyone can actually put such a chart together.
Anthropic hasn’t forgotten about Claude
Perhaps the most intriguing news of all in the last few days is that Anthropic plans to release Claude 4 soon. Like GPT-5, Claude 4 is expected to be a single chatbot that includes quick responses like Claud 3.5 while also handing reasoning queries that are more resource-intensive.
Claude 3 arrived last March, followed by Claude 3.5 in June, so the next major model update is due. Stay tuned for an official announcement.
More
In other AI news, OpenAI’s board formally rejected Elon Musk’s bid to buy the business for $97.4 billion; Mira Murati’s startup has hired another OpenAI cofounder; France’s Mistral has released an Arabic language and culture-focused regional model; AI search-focused Perplexity has shamelessly released Deep Research (no relation to OpenAI’s similar product of the same name); and Thompson Reuters won the first major AI copyright case in the United States — somewhat ironically against legal AI startup Ross Intelligence.
Taking this full circle, Grok 3 is trained on “all court cases” as we’ll probably see tonight.
More on the latest in AI developments in the next edition of Neural — only on 9to5Mac! Read the previous issue here.
FTC: We use income earning auto affiliate links. More.
