I’m using Gemini 2.0 Experimental Advanced on my personal account, and I’ll admit it up front: Ragu can’t wait for this model to be available through the API. If anyone has a connect at Google, we’d take it.
Some context on what this is. Google switched it on for Gemini Advanced subscribers on December 17, and 9to5Google reports that the entry labeled “2.0 Experimental Advanced” in the model picker is Gemini-Exp-1206, which Google says offers “significantly improved performance on complex tasks such as coding, math, reasoning and instruction following.” The picker shows up on the web, not in the app. Google calls it an early preview that “might not work as expected”, says it has no access to real-time information, and warns that it “won’t be compatible with some Gemini features in its experimental state.”
It has a capacious context window. I’ve heard 2 million tokens, but that may be a rumor. The number Google has actually published is for the previous generation: Gemini Advanced running 1.5 Pro got “a greatly expanded context window starting at 1 million tokens — the longest of any widely available consumer chatbot in the world” back in May. The 2.0 announcement, at least as reported, doesn’t put a number on it. Whatever the figure, it is large enough that you can draft your entire article set in one session — if you don’t mind doing it manually.
The API is the sore point. On December 6, Google’s Jeff Dean announced the same model “in Google AI Studio and the Gemini API” — as an experiment. Google’s own documentation says experimental models will typically “not be suitable for production use and come with more restrictive rate limits”, with endpoints that are “subject to change.” You can try it there. You can’t run a business on it.
I think it’s only available on personal accounts. So on Monday I’ll be authorizing everyone in my company to expense the $20 — $19.99 a month for Google One AI Premium — so they can actually use it.
Worth it.
