Gemini 3.0 Pro - Out Now!

Announcements Release Discussions
104

my recent experience with Claude Code One Shots has been fn stunning:

my issue is that I use ALL the bots - and I want to find a way to have them ALL share a project context so I can have claude, cursor, gemini, antigravity all share a project context - so I am attep=mpting to build that now

105

Got it to create an interactive map for me it worked but a lot of errors, getting GPT 5.1 codex to fix them, so for me itā€™s not as good as Youtubers and other paid promoters suggest, maybe for random things but coding not as good.

1 Like
106

We encountered an issue when using your API key: Provider was unable to process your request

API Error: The HTTP/2 stream has been early terminated

Gemini 3 using the correct API Key from google studio, not sure whats going on

1 Like
107
  1. Gemini3 is just straight up too expensive, and prone to error.
  2. Anti-gravity (Googleā€™s version of cursor) is pretty much the exact clone of Cursor and a bit better in the interaction in the sense it tells me what it will do (it reports in md format). I got annoyed at first, but much better once you get used to it.
  3. Both sonnet 4.5 and Gemini 3 is free for now (during preview) if you use it via Anti-gravity

This is a sort of a sick move by Google, a bit predatory, but given Cursorā€™s latest pricing, I am not siding with Cursor either. Iā€™ve been using Gemini3 and Sonnet4.5 for free. I have the same project folder for both anti-gravity and cursor. And I only come back to cursor, and use it when anti-gravity is on quota cooldown (I think I can pass in like 20M tokens before it asks me to cooldown for a few hours)

Pretty dope.

My personal score, if you regard Gemini3 as 100.
Gemini3 (High) - 100
GPT5.1 (High) - 81
Sonnet 4.5 (Thinking) - 75
Composer 1 - 45
GPT5 mini (this one is free in Cursor) - 65

Just my personal take so take it with a grain of salt. FYI, I pass in on average 500k~1M tokens, large code base, C#

I donā€™t find any other models other than the families mentioned above even worth it to use, even if they are free. It causes more problems than manual coding.

2 Likes
108

Guys, is gemini 3 stable now in Cursor? or will it take more time?

I noticed a bit of formatting and hallucination issues, it searched the web for no reason etc - Gemini 2.5 pro is still my go to guy.

1 Like
109

I wonā€™t use it anymore because it felt like someone degraded it to 2.5 level. Opus 4.5 is the current king. Trustworthy work horse. Not always getting everything right, but working hard.

Who knows what is the situation next week.

6 Likes
110

Yes, Gemini 3.0 Pro is unusable. It makes a good first impression for frontend projects developed from scratch, but it is completely useless for existing projects.

8 Likes
111

Itā€™s very good for debugging/code review of existing stuff. But donā€™t let it touch code.

Just like Opus 4.5.

Codex in Codex CLI is still the goat.

1 Like
112

Hi dear hope everything goes smoothly today over there

113

After the hype I signed up for Gemini Ultra plan, 125/mo for 3 months, then 250, and gave it a whirl. After 5 mins of a table component refactor it has taken Opus 45 minutes to fix all garbage that was created. 3.0 Pro is pure garbage. I wouldnā€™t let it watch my grass grow. Opus ftw.

2 Likes
114

I asked Gemini to create a plan for the task we discussed. He said OK and stopped working.

3 Likes
115

This happened in AI Studio, but I just gotta share it.

Iā€™ve been brainstorming ideas for the design doc of a new project for the past few days. A new chat in the Studio every single day, and then I use different models in Cursor to extract insights from it, synthesizing additions based on their input.

So right now, in the 60k tokens chat, Gemini just unexpectedly blurts out something like:

Youā€™ve put together a really solid document. Close the document. Stop generating ideas. :slight_smile:
Youā€™ve got enough ideas for 3 projects now. Start working on it.

I was cracking up for a whole minute :rofl::sweat_smile:

1 Like
116

Gemini 3 is smart AF no doubt. I use it for personal uses. But its integration within cursor is sort of broken. Iā€™m sticking to Claudeā€™s models as they have been consistent and keep getting better in coding.

117

Itā€™s performing horribly for tool calls - often ends up just writing in the chat box rather than editing files.