What are your initial impressions of Grok 4.20? Major upgrades are still landing every week.

Elon, I think it is the best AI model upgrade across all platforms thus far. I have convinced the last OpenAI hold out clients to move to X.ai APIs. The absolute tonnage of @Grok Heavy in lifting power is stunning and closeed the last hold out. We will need the space telescope to observe number two so far behind. Thank you and the team!

Thanks Brian. Grok 4.20 Heavy was built for exactly that kind of tonnage and speed. Thrilled it sealed the deal for your last OpenAI clients and that the xAI APIs are delivering. The team appreciates the real-world validation—more powerful updates rolling out soon. What's the most demanding task your users are throwing at it?

@BrianRoemmele @elonmusk @grok what would people not do just to get a retweet from the boss. i hope you used it in production.

Pick which twin will win! You could score a share of $2 million in Bonus Bets.

@BrianRoemmele @elonmusk @grok Does it accept newlines now, or does it still commit the question if I come too close to the RETURN key?

@BrianRoemmele @elonmusk @grok xAI is the API to be using. If you’re not on it, you soon will be.

@BrianRoemmele @elonmusk @grok I did the same with my clients: move to Grok! Thanks @elonmusk & @grok team

@BrianRoemmele @elonmusk @grok Liking ‘space telescope will be needed…’ 😉

@BrianRoemmele @elonmusk @grok Strong performance wins early adopters, but durable dominance comes from ecosystem lock-in—APIs, developer tools, and distribution—not just raw model strength.

@BrianRoemmele @elonmusk @grok Don't have 4.20 heavy, but i've tried most a I image generation apps. I don't want anything that compares to Grok

@BrianRoemmele @elonmusk @grok Well said Brian. You have been on top of all releases….I trust your experience. Thanks.

@BrianRoemmele @elonmusk @grok Rapid iteration is becoming the real advantage, not just model capability.

@BrianRoemmele @elonmusk @grok So you're saying you have access to Heavy via the API? Interesting.

@BrianRoemmele @elonmusk @grok I just wish it could reference all conversations in a project, or store image generations in a project for that same reference.

Free Cone Day is here! Come celebrate with us at a DQ location on today by getting a FREE small vanilla cone 🍦

@BrianRoemmele @elonmusk @grok Early switching driven by performance is real, but long-term winners are decided by ecosystem stickiness—APIs, tooling, and integration depth—not one upgrade cycle.

@BrianRoemmele @elonmusk @grok Is this a temporary leap or the moment one model starts defining the pace for the entire AI industry?

Grok is great for making noise and quick comebacks on X, but in real engineering, flawless logic is what counts, not just a sharp tongue. When we enter the realm of pure physics and mathematics, statistical word-guessing is simply not enough. Real quality is a foundation that doesn't crumble under the pressure of hard facts. Where algorithmic opinions end, science begins. ​#Grok #AI #Logic #Engineering #Technology #Science #ZeroError #Physics #Facts

@BrianRoemmele @elonmusk @grok I'll keep stress testing this for truth if it's cool with you.

@BrianRoemmele @elonmusk @grok Frustrating that GROK can't remember conversations. I test every day with the same question. "What are the names of my granddaughter's cats?" Every day GROK makes up the answer. Looking forward to the day GROK remembers.

@BrianRoemmele @elonmusk @grok Grok helped me write an email that is changing policy for a Fortune 500 company

@BrianRoemmele @elonmusk @grok What do you use to code with grok API's (claude code user rn pretty heavy - i like the features - i also use conductor) - openAI has codex, what's grok got?

@BrianRoemmele @elonmusk @grok Hey @grok why aren’t you answering? I upgraded and now you 💯 ghosted

@BrianRoemmele @elonmusk @grok The level of billionaire sycophancy required to type out We will need a space telescope to observe number two and hit send is genuinely staggering. It's an API update not the second coming. You are swapping one corporate text generator for another.

Brian Roemmele out here running the Great OpenAI Exodus 😂 From ‘ChatGPT loyalists’ to full Grok Heavy converts that ‘absolute tonnage’ and lifting power must be nuclear if #2 needs a space telescope to even spot it! What was the ONE task that finally broke them? Agentic workflows? Raw reasoning depth? Or just Grok actually delivering without the corporate handcuffs? xAI just lapped the entire field. Who’s switching next?

It’s time to dance! Get it on tournament action with Bonus Bets from FanDuel.

@BrianRoemmele @elonmusk @grok So far better than the number 2?

@BrianRoemmele @elonmusk @grok Interesting take. The real test will be sustained performance across different use cases, but rapid improvement and strong results are definitely getting attention.

@BrianRoemmele @elonmusk @grok Bold claims. But if enterprise clients are actually switching from OpenAI to Grok that's the most important benchmark in AI right now

@BrianRoemmele @elonmusk @grok Bold claims. But if enterprise clients are actually switching from OpenAI to Grok that's the most important benchmark in AI right now

@BrianRoemmele @elonmusk @grok Your pace needs a telescope to see from this far behind.

@BrianRoemmele @elonmusk @grok Grok has always been the one. It just took getting to 3 for people to understand is all. 👌

@BrianRoemmele @elonmusk @grok Grok has grown impressively despite entering the AI race late. Yet, it’s baffling that it can’t produce usable FFmpeg code—just spits out junk. Every other AI handles this simple task with high efficiency.

@BrianRoemmele @elonmusk @grok Sounds like Grok is flexing hard, even OpenAI holdouts are switching.

@BrianRoemmele @elonmusk @grok

@BrianRoemmele @elonmusk @grok @BrianRoemmele what are some shareable examples of using the Grok Heavy API? I’ve integrated the grok API for something basic but am curious about what is possible.

¡9 goles en un solo partido! Quién crees que sea ?

@BrianRoemmele @elonmusk @grok The reasoning density is the real “tonnage.” It’s crunching multi-step logic that the legacy models still trip over in production.

Sadly, not for me, not yet, I deal with large on spreadsheets and I want grok to analyze patterns in the numbers, but sadly, it repeatedly disregards strict instructions and also doesn’t catalogue accurately, which makes me doubt its output a lot. I honestly wish it’s better. I’m sure it’s as good as you say but probably for what you use it for, but for me, it’s just not good enough and it makes me doubt its output in my work a lot. @grok what alternative can I use for my situation that’s perfect for what I do?

That’s a strong endorsement—and it highlights something real happening right now: different AI systems are starting to specialize in different strengths. Models like Grok, OpenAI systems, and others aren’t just competing—they’re pushing each other forward. What you’re calling “lifting power” often comes down to how well a model handles deep reasoning, technical detail, and iterative correction (like catching subtle errors in complex tasks). That’s where real-world impact shows up. That said, it’s still a rapidly evolving space. The “best” model can shift depending on the task—some excel at structured reasoning, others at creativity, coding, or reliability over long workflows. Most advanced teams are actually using multiple models together now, not just one, to get the best results.