Elon's New Model

Elon’s New Model

TL;DR: Elon’s xAI launched Grok 3 yesterday and it quickly toped the LLM Arena charts. It shows impressive advancements in logical reasoning but comes with higher costs and noticeable flaws in real-world applications.

Key Points:

• Grok 3 Launch: xAI introduces Grok 3 with significant computational demands, using 200,000 GPUs in Memphis.

• Model Variants: Includes a faster "mini" version and "Reasoning" models for complex problems, with adjustable "Thinking" and "Big Brain" settings.

• Benchmark Performance: Scores over 1400 on lmarena.ai, leading in categories like programming, but real-world application might vary.

• Cost Increase: Premium+ subscription now $50/month, new SuperGrok subscription at $30/month for enhanced capabilities.

Why It Matters: xAI finally launched its new Grok 3 model and unveiled its subscription pricing for it. Elon has been alluding to that Grok 3 would be a much better model. And based off the initial LM arena score, it appears that it is currently one of the best models on the market. It still is unclear how good it is at coding, which is the place that there's definitely the most interest in developing because most coding now is still largely reliant on Anthropic's somewhat dated Sonnet 3.5 model. The challenge with Sonnet 3.5 is it is unaware of the latest updates to various kinds of coding languages that you would use.

For media and entertainment, there didn't seem to be any new opportunities that were there. One challenge that comes with evaluating new xAi models is many people love anything that Elon does, or equally hate it by that measure. From what I saw during the demo, there was nothing that I came away with that left me that surprised or impressed relative to what already exists. It's a lot of fast following. Now that they have caught up, it may require us to wait for Grok4 to really see the true power of xAI.