Gemini team has a long history of releasing great benchmark charts and disappearing into the ether. There has never been a moment where I have preferred them to Claude/Codex. When I get access I eval them on my home assistant set, a small coder benchmark, etc. and they are too expensive, and too low quality.
The benefit for Google is their chat interface has privileged access to their data (maps, etc.) but their model itself is low quality and, if I'm being honest, has the promotion-oriented measurable-output bias that FAANGs tend to have. The later Llamas had the same problem.
About to say the same... even with the "Gemini AI Plus" plan that should be renamed to "Gemini Flash" because you are only allowed to use "the mangled model" using it for something as basic as taking notes and saving to google keep or even google docs, heck yeah a simple "add this to my calendar" completly fails in gemini... Gemini sucks for everything even its own google apps... Claude and Codex even DeepSeek and Kimi are milestones better than anything Google has launched.
It is unfortunate that having such an amazing engineers that are capable to solve any leetcode exercise in 5 minutes on the spot with a white board, they aren't able to even deliver something useful on the AI space.
I love this never-ending people having vastly different experiences situation with LLMs. In my experience, Gemini flash 3.8 is an exceptional model and is the bulk of my agentic use as someone who bounces around models as soon as they come out. I would say it’s the best flash model until 1017 came out. It might still be better than 1017.
On the free and "AI Plus" plan you are only allowed 3.6, 3.1 Pro and flash "lite" (3.5), 3.8 is for the Pro users (the ones paying the full membership)...
Daily usage: normal task, modify this document, do this do that basic office task... anything that requires precise context is doomed for any google-based llm, that is a well known issue of even the gemma models. Only time it has been useful is while using antigravity that makes sure everything has context on every turn therefore wastes more tokens than Fable on the same prompt. Any other LLM will do it faster, burning less tokens and with better results.
Bizarre to me as well, but it's obviously a use case difference.. I wonder what they are doing? I have found it to be excellent for my tasks (reverse engineering 6809, SH-2, FPGAs, etc).