I was a big fan of Horizon Alpha, a stealth model available on OpenRouter, later revealed to be an early checkpoint/variant of GPT-5. Unfortunately, the release candidate isn’t quite as good at my usual vibes benchmark when it comes to creative writing, be it 5 or 5-Thinking.
(4.1 was really good at the same task, and surprisingly so, for a model marketed for coding. I missed it when it was gone, and I’m glad to have it back)
I was initially quite negative about GPT-5T, but I’ve warmed to it. It seems as smart or smarter than o3, and is a well-rounded and capable model. The claims of a drastically reduced hallucination rate seems borne out in extensive use.
I was a big fan of Horizon Alpha, a stealth model available on OpenRouter, later revealed to be an early checkpoint/variant of GPT-5. Unfortunately, the release candidate isn’t quite as good at my usual vibes benchmark when it comes to creative writing, be it 5 or 5-Thinking.
(4.1 was really good at the same task, and surprisingly so, for a model marketed for coding. I missed it when it was gone, and I’m glad to have it back)
I was initially quite negative about GPT-5T, but I’ve warmed to it. It seems as smart or smarter than o3, and is a well-rounded and capable model. The claims of a drastically reduced hallucination rate seems borne out in extensive use.