SpaceXAI just launched Grok 4.5 to the public on July 9, and the first thing you notice is not the benchmarks. It is that SpaceXAI is playing a completely different game than it was three months ago.

Grok 4.5 is an Opus-class model trained on a 1.5 trillion parameter foundation built with Cursor coding data. That last part matters. Unlike GPT-4o or Claude Opus 4.8, which were trained on broad internet data, Grok 4.5 was purpose-built for code because that is where the margin is right now. A coder using this wants reasoning and speed and low latency. They do not care about your latest benchmark on some abstract reasoning task.

The pricing confirms it. Standard tier is $2 per million input and $6 per output. Premium is $4 and $18. That undercuts Claude Sonnet 5 ($2/$10) on the standard tier and sits in the middle for premium. It is not the cheapest. It is not trying to be.

What struck me reading the launches is what Elon said: Opus-class, but faster, more token-efficient and lower cost. That is the codex for 2026. Not who has the best model anymore. Who can deliver Opus-level reasoning at Sonnet-level economics.

Grok 4.5 beats Opus 4.8 on half of xAI’s published benchmarks. Loses on the other half. That is fine. It tells you the thing is actually competitive, not marketing fiction.

The real signal is that xAI is releasing a new foundation model every month through the end of 2026. That is aggression. That is a lab that thinks it can close the gap on reasoning and is willing to burn compute to prove it. Whether it works at scale is a different question, but the intent is clear.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://eciks.org/12495-68037-grok-4-5-spacexai-launch-opus-class
https://explainx.ai/blog/grok-4-5-public-launch-spacexai-july-2026
https://www.analyticsinsight.net/news/spacexai-launches-grok-4-5-ahead-of-gpt-56-race-what-we-know-so-far

Repost this. Thanks.