3.5 Flash is way better than Gemini 3.1 Pro. Unbelievable, it is so fast and accurate. This should be 3.5 Pro while the upcoming 3.5 Pro should be called 3.5 Ultra. It's time to retire the Flash name and move up the barrier. Just Pro and Ultra models.
I used 3.5 Flash to tackle a bug in Fibre where MPMC would leak memory due to the way it handles and doesn't handle adding and removing waiters quickly. The ways that 3.1 Pro and Claude did always came up with a regression in performance 50% - 75% drop in benchmarks, but the correct fix. Obviously I rejected it. 3.5 Flash, on the other hand, with some guidance on where I wanted to go with regard to using state machine came up with an unsafe way to implement it and boom, I gave it to Claude Code, did tests, benchmarks, miri, etc. A few rounds of these to fix UB issues and we got a solution that works as fast or faster than before the fix. That's that kind of solution I expect from a Pro model.
3.5 Flash can handle complex systems programming tasks, that's PHD level work. The talk of AI being a junior developer is flat and experiences of people who do not use AI.
2.5 Pro was the first model where I saw a big opportunity to develop a LOT of software. I'd argue at least 10-20 years of work was done over a period of less than 1 year. Gemini did go through regressions after the first 2.5 Pro, but it looks like a year after with 3.1 Pro and 3.5 Flash, it's back and better.
Use AI for tutoring, interview practice, coding, discussion, everything. The sky is the limit. GO GO GO.
