It's impressive how badly Google has screwed up their models and coding products. There was a brief period of time I was using Anti-gravity as my daily driver when they first introduced it and then they just completely screwed it up.
Same. The reason I gave up was their limit inconsistency. I signed up for their cheapest sub and got what I felt was fair. I subbed for ultra and hammered the opus as much as I could - first week I couldnt hit the 5h limit at all, used 80% of weekly limit. Second week I couldnt hit the 5h limit at all, used up my weekly limit 1 day before reset.
Third week I was hitting 5h limits by hour 3, weekly limit in 4 days. Last week Ive hit my 5h limit in under 40 min, and stopped using it completely.
My usage was very consistent thorough the month. I wasnt launching subagents, I wasnt working with more full caches, they just tightened my limits during the month I paid like 180$ for, which was discounted 50% for first 3 months, so they wanted over 300$ to keep the sub I hit 5h limits before my cofee gets cold. Absolutely insane
Not just their coding products. Compare Claude Desktop / ChatGPT Classic to Gemini Desktop... Gemini desktop is barbaric with less usability / polish and more UI bugs than its online web counterpart.
I wasn't able to give the new ChatGPT desktop app a try because it insists on phone number for the registration flow even though I'm using a free tier.
Erm, that's to be expected. GPUs ain't free. Anyone can make a free account and cost them potentially infinite compute, so a phone number is a way to lock that down. Might not be the "best" way, but hey: it's free.
I suffered with Antigravity 1 and 2 for a long time (I’m not some weirdo, I just run a blog where I do deep dives into AI tools for programming=). The desktop version was basically unusable, so I used the console one. A couple of weeks ago I finally dropped it and switched to Pi with antigr login. With plugins you can configure it however you want. I just installed one of a deep-search plugins too: you need to add API credits from a search engine, but it can run multiple search rounds with a specified depth.
Compared $20 ChatGPT and AGY plan: my weekly limit on sol/terra used to disappear in one intensive day (or two normal ones). Now they’ve added the 5-hour limit on top of that, which is even more annoying. With Google I barely notice the limits at all — the only one is Opus 4.6 (not sure why it has one), and I only use it for double-checking sometimes.
The biggest problem is that the Pro model hasn’t been updated, and Flash just doesn’t produce solutions of the same quality as Sol. But it is matter of time...
I had the same poor initial experience that others reported, but has anyone been paying for it recently and using it as a serious tool? Their new models are a pretty good value and crazy fast; I wonder if the Gemini on Antigravity output is overall competitive or has some niche use case other than being a frontend for Agent Substrate and their other projects
Got off Gemini before they ended the cli, was not working for me, and went to Codex and opencode/deepseek flash v4 0731 which was working ok.
But got the google ai pro free 1 yr student - been liking agy quite alot actually - pretty good weekly allotment. Like the default, Flash 3.7 high.
Even bought extra credits I've been hammering it all day long.
People talk all day about coding but where this thing amazes me is devops. It hooked itself to discord so i can command it, set up or troubleshoot ssh servers on my home network. Blew my mind a couple times.
I use it for all my non-professional work and I find it works well enough for me.
Mostly for things that I know are possible and have a vague idea of how I would do it, but it would take me longer. Some examples in no particular order:
* Made my wife's travel blog/site
* Chrome extension for Crunchyroll to allow multisubs + local translation for unsupported languages (so my wife and I can watch and she doesn't need to focus extra hard reading English)
* Same as above for Netflix
* Setup multi subs with mpv (some Lua plugin shenanigans)
* Setup proton for warcraft 3 + wc3 champions (unofficial ladder) on Linux
* Setup a bunch of xrandr aliases for my different monitor uses (I'm still on x11)
If you need a custom chrome/firefox extension or to create some bespoke Linux QoL stuff, it works great. This was all Gemini Flash 3.5-3.7 (been using it for a while). I suspect pretty much anything would succeed at these tasks, but I just stick with antigravity because it works for me(and I'm on some discounted plan).
My experience recently is that antigravity CLI is much better than Gemini CLI now. Gemini would often get stuck in loops before. Fine for personal projects at least, especially for the price (got a year’s sub for $5 on G2A)
But I don’t think it’s as good as Claude Code or Codex yet.
Am also curious on this! Though if it’s great now I’d need a sustained period of consistency (let’s say six months) before I try it again - the rollercoaster of getting a new randomized feature and quality experience every week got old pretty quick. (And I felt bad for recommending it to a few people at the time based on a good week)
My experience with Antigravity is that it will tell me it's completed a task or implementation that it hasn't. Like, the work is half-mostly done but not.
Judging by this entire page being the direct output of an LLM with little editing done at all I assume the feature is probably half baked and probably just a bunch of skills or harness tweaks? In that case, why would anyone pay for this at all?
As someone on a google workspace plan, where the monthly price went up to include AI features, I’m disappointed that google won’t let me sign up for antigravity development plans with my workspace google ID. To the best of my knowledge, as of last week, the recommendation is to sign up with a different, personal google id.
It feels like such a stereotypical big corp misstep, but based on the reputation that google’s AI products have, I guess I’m not _that_ heartbroken. It’s mostly frustration that they forced the price increase on everyone, and I’m definitely not receiving the benefit.
On the one hand this kind of harness work is pretty interesting, on the other the blog post is light on details and sounds like existing behavior of any high end model in Claude Code, Pi, or one of the litany of other projects that people seem to be cooking up on a daily basis.
> Google Antigravity's Editor view offers tab autocompletion, natural language code commands, and a configurable, and context-aware configurable agent.
They’ve had this weird phrasing on this landing page for at least 6 months if not longer. It’s clearly a landing page that receives no love.
And yet I just installed Antigravity, and it had a default checked box to also install Antigravity IDE. So someone still cares enough to try and juice the install numbers
It's because they did a very bad rollout with 2.0 which totally removed the editor, which frustrated a lot of people, so then they added all these ways to install the version with the editor again to try to fix that.
Their attention is definitely focused on the non IDE tool
I canceled my Ultra subscription after realizing I had Sol constantly fixing anything I tried to do with Antigravity. The most frustrating experience and a waste of money. I’m sorry for Google and hope they manage to fix their disaster.
I have some super specialized workflow I wish I can run with their model, but looks like they don't allow programmatic call (claude code and codex allows it) in their plan.
What should you be looking out for as bad behavior from this model? I've daily-driven 3.5, 3.6, and now 3.7 for however long it's been out (alongside Opus, as I hit daily limits on both subs each day) and haven't noticed anything "untrustworthy".
My usage was very consistent thorough the month. I wasnt launching subagents, I wasnt working with more full caches, they just tightened my limits during the month I paid like 180$ for, which was discounted 50% for first 3 months, so they wanted over 300$ to keep the sub I hit 5h limits before my cofee gets cold. Absolutely insane
If you give it a task, it completes it. It might so a single agent chug, or delegate some parallel tasks..
The architecture is essentially what was agreed on during the planning and spec phases.
So I'm just not understanding like... The codebot produces everything I ask, I use Gemini all day every day and I don't hit limits.
I see that if I were trying to code 10 hours a day the limits might be a problem but... This is a personal plan.
So. When people say that Gemini sucks, or they're hitting limits, I'm not understanding what problems they're encountering...
I wasn't able to give the new ChatGPT desktop app a try because it insists on phone number for the registration flow even though I'm using a free tier.
Erm, that's to be expected. GPUs ain't free. Anyone can make a free account and cost them potentially infinite compute, so a phone number is a way to lock that down. Might not be the "best" way, but hey: it's free.
Compared $20 ChatGPT and AGY plan: my weekly limit on sol/terra used to disappear in one intensive day (or two normal ones). Now they’ve added the 5-hour limit on top of that, which is even more annoying. With Google I barely notice the limits at all — the only one is Opus 4.6 (not sure why it has one), and I only use it for double-checking sometimes.
The biggest problem is that the Pro model hasn’t been updated, and Flash just doesn’t produce solutions of the same quality as Sol. But it is matter of time...
ultracode is different to ultrathink (because of course it is...)
ultrathink was "throw more tokens" (it was a Claude feature that has gone away[1])
ultracode is "throw more tokens AND write code that creates and orchastrates subagents": https://code.claude.com/docs/en/workflows
[1] https://claudelog.com/faqs/what-is-ultrathink/ claims it has returned, but that might not be accurate
But got the google ai pro free 1 yr student - been liking agy quite alot actually - pretty good weekly allotment. Like the default, Flash 3.7 high.
Even bought extra credits I've been hammering it all day long.
People talk all day about coding but where this thing amazes me is devops. It hooked itself to discord so i can command it, set up or troubleshoot ssh servers on my home network. Blew my mind a couple times.
Mostly for things that I know are possible and have a vague idea of how I would do it, but it would take me longer. Some examples in no particular order:
* Made my wife's travel blog/site
* Chrome extension for Crunchyroll to allow multisubs + local translation for unsupported languages (so my wife and I can watch and she doesn't need to focus extra hard reading English)
* Same as above for Netflix
* Setup multi subs with mpv (some Lua plugin shenanigans)
* Setup proton for warcraft 3 + wc3 champions (unofficial ladder) on Linux
* Setup a bunch of xrandr aliases for my different monitor uses (I'm still on x11)
If you need a custom chrome/firefox extension or to create some bespoke Linux QoL stuff, it works great. This was all Gemini Flash 3.5-3.7 (been using it for a while). I suspect pretty much anything would succeed at these tasks, but I just stick with antigravity because it works for me(and I'm on some discounted plan).
But I don’t think it’s as good as Claude Code or Codex yet.
Have they ever gotten around to fixing it or is Antigravity only feasible with their Ultra plan?
It feels like such a stereotypical big corp misstep, but based on the reputation that google’s AI products have, I guess I’m not _that_ heartbroken. It’s mostly frustration that they forced the price increase on everyone, and I’m definitely not receiving the benefit.
They are ignoring antigravity IDE like it does not exist anymore.
https://antigravity.google/product/antigravity-ide
They’ve had this weird phrasing on this landing page for at least 6 months if not longer. It’s clearly a landing page that receives no love.
Their attention is definitely focused on the non IDE tool
For real though, I think their reputation with developers is in the gutter after years of trial balloons.
Every single hour on HN it's OpenAI does this, Gemini does that.
It's all nothing.
Extra reasoning. Big whoop.
the primary difference between the free and the paid versions should be the amount of inference quota you have access to
3.7 is very reliable.