This model was scheduled for release like 1 month ago, what happened? Is it ever going to be released? Any timeline users can look forward to? At the moment the difference between gemini models and Chinese OSS/OpenAI/Anthropic models is so big that it really feels hard not to ask any questions.
Probably in next two weeks
Gonna be outsider for real
I don’t think they can release anything that doesn’t compete with the latest frontier models. What would be the point? I don’t work there, but it feels like the vibe is overrun by corporate bean counters and too much BS for employees to deal with. Why wouldn’t you want to jump ship and go somewhere you feel supported in the AI work you want to do?
No idea but it’s just really weird because it’s been SOOO long since the release of gemini 3 pro, essentially very little if any progress after that. I feel that a company as big as google could have and should have made more progress. What is happening is not normal, AI is not some mystic thing anymore. People have been studying how AI works for years by now, it’s not as hard to find engineers that can create new competitive AI models.. Or they could just host Kimi k3 if they really can’t come up with a decent pro model themselves …
The most frustrating part about the delay is there is no access to a deeper thinking reasoning model. It is either flash 3.5 lite or flash 3.6 unless you want to switch to pro which is only 3.1 and getting very outdated. The flash models are fast and capable for a lot of things, but honestly I would prefer to still have Gemini 3.0 thinking was better than everything that’s available right now.The only model that has any type of actual thinking and logic is 3.1 Pro and that model has already been known to be somewhat restrictive due to them creating it in order to pass certain benchmarks. It is very stiff and rigid. Whereas with 3.0 thinking it was not the pro version but it still actually thought about and processed your prompt and gave you a good answer back because not everything should be about speed.It would be nice if they had legacy model options like ChatGPT gives if we still want to use 3.0 thinking instead of 3.5 flash or 3.6 flash because they’re only good when you need to do something quickly but they don’t reason well enough and hallucinate a lot. Coming from a long time Google AI pro subscriber I don’t plan on canceling and I get a lot of use out of my subscription for a lot of the other things that are in the ecosystem within Google like Flow and NotebookLM or Gemini LM now but we need a thinking model that’s not pro 3.1
I’ve been triying Codex for the past 2 months on a 20 dollars subscription. I got more done with Codex in 2 months with this sub than when I had subscription for google for 1 year. Gemini models are 1 prompt decent answer, next 5 prompts hallucination like there’s no tomorrow, then 1 more decent answer mixed in there randomly, then hallucination, then bug, then prompt again and it’s hallucination loop.
OpenAI releases a new flagship model every 2-3 months. Google has released 0 flagship models in the last 8 months, and even 8 months ago they were quite behind anthropic and OpenAI. Now even the chinese models like Qwen 3.8 max and Kimi K3 are far superior to Gemini models, like not by a little bit, but by a MASSIVE amount.
I don’t know how companies with 1/1000000000000000 the money that google has and with nearly no infrastructure, and with 1/100000000 of google’s engineers count, manage to create such massively better models than google.
It’s genuinely crazy to me that there are people still paying for a google subscription under the current situation. Furthermore, even when the had just released 3.0 pro, with antigravity rate limits it’s impossible to build anything unless you pay 100+ dollars, which is absurd for most normal humans to go and pay 100 dollars for AI subscription.
I hope at least that they are making progress with medical AI models that can enable research that was impossible to do before, because my only hope is that they might have been busy with that, and therefore did not manage to output a new pro model every 2-3 months, otherwise it’s just beyond shocking.
Codex is really good. I allow mine to control my Google SDK and Gemini CLI as well as my Google cloud account and it is able to create custom wrappers that use my Google cloud free credits. Google cloud and all their different APIs and ADC logins with IAM credentials is confusing. But when I allow Codex 5.6 SOL to access everything it can map the proper SKU to hit the correct bucket of credits.