Grok-2 is here, and it's gunning for ChatGPT's throne.

xAI just dropped the mic with Grok-2, their latest AI model. This one isn’t just about Musk’s gimmicks of anti-woke or slur-filled roasts. It’s got some serious punch (GPT-4 class) in it. Grok on Twitter has also got some UI updates and image generation features.

What's going on here?

xAI released Grok-2 and Grok-2 mini, new AI models available on X and soon via API.

What does this mean?

Grok-2 joins the weight class of GPT-4o, Claude 3.5 sonnet and Gemini 1.5 Pro. It’s quite not as good as them, but xAI says that this is a beta release. It’s reflected in its ranking (#4) on LMSYS leaderboard as well. I expect them to make similar incremental updates and go for that #1 position for at least a few days when they do a general release.

Grok on X (available to Premium and Premium+ users) get first dibs on Grok-2 mini right now. Grok-2 is coming soon. Both of the models are going to be available to developers as enterprise APIs too.

Along with these models, the interface for Grok on X is revamped. They're also experimenting with image generation using the new Flux models from Black Forest Labs. Looks like the partnership with Midjourney hasn’t quite worked out (yet).

Why should I care?

In the baby tests I have been able to do: The model (even the mini) is powerful but integration with Twitter is still very weak. It gets random tweets as sources and then tries to combine them into a coherent answer. So the claim of “use Grok with real-time info on X” is a bit sus (just like its test name on LMSYS: sus-column-r 😜)

If you used to switch tabs to go to ChatGPT when using Twitter (for whatever reason) Grok is a good replacement now. Just don’t expect anything extra because it’s within the X platform—the integrations have much to catch up.