Notes on Flash 3.8

Google released Flash 3.8, a new version of its Gemini Flash model, available in Antigravity. In an informal review, Nate Meyvis found that Flash 3.8 with low thinking was about 25% faster than Flash 3.7 on a content-extraction task, but results were not clearly better or worse, and high thinking mode was slow.

Flash 3.8 https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/ is out, as I discovered when it showed up in my Antigravity https://antigravity.google/ terminal. A few notes after perhaps an hour of use: - As I discussed earlier today https://www.natemeyvis.com/first-impressions-of-fable-51/ , I use the latest Flash models heavily because they combine power and speed in such a valuable way. - Flash 3.8 with high thinking did well on the first tough task I gave it, but it was slow probably not faster than Fable . I'm likely to keep using Fable 5.1 instead of Flash for tougher tasks. - Flash 3.8 with low thinking is addressing simpler tasks, both coding and non-coding, quickly and accurately. I believe that it's more powerful than 3.7, but I haven't yet felt that difference. - So far, there's still a big latency difference between low and high thinking. - I have a standard content-extraction task that I like to use as a benchmark. Over a small sample, Flash 3.8 with low thinking gave results ~25% faster than Flash 3.7 low thinking; those results are different but, in my judgment, not clearly better or worse. - "Not better, but faster and a bit different" fits my experience across the ten or so little projects I've done with it so far. Again, given how I use Flash and how important speed is, it would be great if this were true. Happy New Flash Day