Grok 1.5 - xAI's new model reasons better

Elon’s xAI has been making progress quick now. It open-sourced Grok-1 less than 2 weeks ago and is now announcing Grok-1.5.

What’s going on here?

xAI announced Grok-1.5 and improvement over their previous model Grok-1.

What does that mean?

Grok 1.5 gets two improvements over it’s baby brother Grok-1.

• It has gotten better in reasoning, math, and code generation.

• It joins the long-context game with accepting upto 128k tokens.

Based on benchmarks, Grok-1.5 looks like another model that’s almost near GPT-4 level (March 2023 one) but not quite there. Others in this category are Mistral-large and Claude 3 Sonnet.

Grok-1.5 will be available to early testers and users on X (Twitter) soon.

Why should I care?

Although there’s no mention of open-sourcing Grok-1.5, I consider the possibility high based on Elon’s ongoing lawsuit with OpenAI. It hasn’t been long since Grok-1 was open sourced, so we have to wait and see how it affects the open-source ecosystem. But I consider that it won’t negligible.

Also, it’s interesting to see xAI making quick progress. Grok-1.5’s context window should be able to use near a thousand tweets in one go. Imagine asking Grok, “Hey, what all did Ben tweet about in last month?”