Last week in AI: GPT-5.6, Grok 4.5, and usage resets for everyone.

Hey there,

What a busy week. The logjam of model releases finally cleared and the tokens are flowing.

OpenAI launched the GPT‑5.6 family of models. We already knew these were coming, but they seem even better than expected. Evals broadly show it doing as well as Anthropic's Fable 5 but with fewer tokens — critical as the recent news cycle was obsessed with how expensive tokens are getting in the wake of Fable's release. For more, read Every.to's review.
Fable is gone and then back and now 50% off (until tomorrow). The date Fable leaves the Claude subscription tier has been pushed back twice while the usage has been discounted heavily. Safe to say launches from other leading labs had an impact here. Capitalism is doing it's thing.
X.ai released Grok 4.5 in collaboration with Cursor and it's really good. Cursor's benchmark puts it at near-Fable performance at a fraction of the cost. For the people trying to code on a $20/mo plan, this model probably makes the most sense as your daily driver.

One update from me:

I ran salesevals.com against GPT 5.6 and it did great. I was't too surprised, GPT models were ahead in the benchmarks to start with. The most surprising thing: all three max models did about the same with Luna being 6x cheaper. Updating most of my workloads to a 5.6 model this week.

As for next week, my gut tells me we'll see more model chicken a usage limit resets. Ironically, we'll also see how the chip super cycle is doing: ASML + TSMC both report earnings. And we also get the CPI inflation numbers while the Strait of Hormuz is closed again. So much for a slow summer vacation week.

LFG,

Drew

Don’t miss the next one.