So I’ve used Antigravity as ran out of my budget this month - its slow as hell.
Even the so called “Flash” model takes ages to do anything.
Anyway, I’m not sure what the point of this post is but Composer-1’s a fantastic model and to be honest well-worth the money. Props to developers of it.
ya, yesterday i tried this composer 1
it displayed in dashboard as free
is there any dicount or promo now?
yeah, I’ve been using the composer-1 for a while basic only using auto + composer-1 for 1 month already
Antigravity is way faster and better than Cursor, for me. The only issue is that MCP servers are buggy. Antigravity is immensely cheaper than Cursor.
I am surprised, maybe I’ll give it another go.
Agree on the Playwright ones, its hopeless with it. I swear Playwright/Chrome MCP used to be a lot better than what we see now and a lot faster. I rarely use it these days because of how slow it is its quicker just to do it myself
Some things may be slower but, in Antigravity, the agent can make reliable edits to multiple files in parallel, which drastically speeds up such operations. Overall, it is much faster, way cheaper, and far more reliable. Their system of creating implementation plans, task lists, and follow-throughs and the indexing them for cross-thread memory is also pretty slick, if still not an optimal remember everything system. Bottom line, it is way cheaper. But some mcp servers won’t work in it. Seems to relate to the number of tools they have. Python mcp servers seem more problematic than typescript, but I am not sure.
I think Antigravity has uncertain future, Google has a tendency of discontinuing their products. They have more than one ‘agentic IDE’ and I think it is a bad practice. It is probably cheaper than Cursor, but I believe this either changes soon or there will be features missing in comparison to Cursor. Also likely Google has rushed the development of Antigravity and might not pay as close attention to it as the company itself is highly sustainable without its existence. Cursor on the other end is defined by its IDE. (although there seem to be way too many issues in recent months 
)
I personally do not want to switch IDEs every other month so I stick with Cursor.
Also Cursor has a great way to get in touch with their team, right here on the forum, which is extremely valuable to me.
Well, after i’ve hit the cap last month after 10 days of usage with auto, I started immediately looking for an alternative. My work pays for the pro subscription of ChatGPT, but their codex agent is still awful (codex 5.2.. So slow and with underwhelming intelligence), in comparison the amazing results of the agents I used with cursor. I also checked out GitHub copliot on vscode, which was ok - but their UI was really annoying.
I checked anitgravitty when it first launched, and the free tier was buggy and ran really quickly out of quoata.
However, after seeing some reviews online, decided to give them a change and have bought the AI-pro subscription (21 euros, 4 more than cursor) - and wow!
The UI is really nice and helpful.
While they don’t offer ask/debug mode, their planning capabilities are really good.
The models are fast!
Most importantly - The quota. It divides into 3 sections:
And they replenish EVERY 5 hours. First of all, you know how much you’re using, so you could plan - no more shooting in the dark, like with cursor. Working smartly (also using chatGPT and Gemini chat online) - I always have enough quota for all my work, and I always wake up to a new day with a fresh quota, which is such a comforting idea.
The unknown feeling that we all have with cursor actually makes me not use it - stupid, I know, but it was frustrating to simply run out of juice in a middle of a hard working session.
Honestly, I wish that cursor would offer the same model - their models are better than the current state of Gemini 3 pro. Opus 4.5 is definitely amazing, though. Especially for planning.
Until cursor decide to be more clear about our usage or fight back Google’s excellent model, i am with anti-gravity. Coming back every day to the forum though, to see maybe there is light in the end of the tunnel…
Since I cancelled my Cursor subscription as I can’t pay $800 per month for per-token agents, I found that even VScode had a fairly reasonable GPT5-mini model that they offered 100% free and without limits, and for most of my tasks even that was enough. Since I am working with C++ code, the underlying complexity is usually so high that even the most expensive models can’t deal with it without making ugly mistakes, so I have to keep an eye on everything the AI does. With such constraints, even a less capable model’s agentic workflow is a great help, it was not better than cursor, but for my use-cases it was not worse either.
Then I tried Antigravity, and I must say it is an absolute Cursor killer. In the past 4 weeks I went from a buggy application to a full product release, with e-commerce implementation, licensing implementation, key generation, automated upgrading, and now I have a product for purchase. Not entirely due to Antigravity, obviously I was also working 24/7 to get there, but it did help a lot.
Finally I had the chance to work with the Opus model, which I could never try under cursor because of the price. I don’t think it is worth what Anthropic or Cursor is asking for it, it is slightly better focussed than other models, but if I keep using the same context for too long it can make very ugly beginner mistakes too. Gemini 3.0 Pro is there at the same level as Opus 4.5, maybe even a slightly better. I would use Gemini 3.0 for longer workflows, Opus for smaller but complicated, deep tasks. For both models it is still very important to stop reusing the same context for too long, as both become unusable after a while.
In my opinion we are already witnessing how the AI bubble is going to burst and in my opinion Google is already beating whole laps on the rest of the participants. They have their own chips, they don’t have to pay NVidia ridiculous premiums, the only area where they use the same resources is when they are buying new RAM. Even with that, they already owned huge data centers before the AI craze started, so they are running with an advantage compared to others. Whether it’s going to be good or better for us in the end, it is difficult to tell, maybe we’ll use adblocker extensions in vscode in the future. For now it is a fact that they are far better than anyone else.
(Just to clarify my costs: I paid $20 for a Google AI Pro subscription, and I have been able to use Gemini 3.0 Pro and Opus 4.5 almost 100% included, without paying a penny extra. I did hit the limit on a couple of occasions either with Opus/Sonnet or with Gemini, when this happened I was less than 1 hour away from the 5 hours limit reset so did not have to wait for too long. But I could also just switch from Gemini to Claude, or Claude to Gemini, and continued the work like that. I was already paying Cursor $60 when they removed the unlimited auto feature, and my usage would have skyrocketed to $800 per months with the ultimate cursor subscription, so $20 is pretty convincing for me.)
Cursor has more features, at least superficially. But they have a lot more bugs. I’d say it’s a pretty even trade-off in that regard depending on your use case(s). But AG is way, way, way cheaper. And it is far faster and massively more reliable making edits to files, especially multiple files. It just works better and cheaper. I suppose Google might dump it at some point, but at this point it is King, at least if you don’t need the most reliable mcp server support.
Where do you find the usage charts from your screenshot?
I am currently double paying for cursor and antigravity fml. Have ya’ll tried opencode + claude code as well?
The quality of code and architecture coming from Antigravity for me has been much stronger. The UX, especially for file review, is excellent.
The buggy mcp (esp supabase) is crippling and the lack of discord/forum is not great.
But working with antigravity’s models feels more like having a collaborator and I am fighting it less on bad practices/junk code, which has been a huge issue for me with cursor.
I love cursor and want them to win, since they originated this category.
But concerned that some of the UX and agent capabilities are lagging.
eg with skills and multi agents Claude Code seems to have a ton of things that would be really useful and not clear when cursor will add something equivalent.
Antigravity right now is better than cursor in stability. I am about to cancel my cursor because they ship the most buggy updates every week! Antigravity way cheaper too.
Cursors change to forcing that stupid “Review” page, instead of just taking me to the freaking file I clicked on to review inline, finally drove me to look elsewhere.
So far I quite like Antigravity, the tab completion seems a little slower then cursor but the review functionality is the same as cursor used to be, so that immediately makes it infinitely better then current cursor to me. It actually feels closer to like the old 1.5/1.7 version of cursor.
I also feel like I’m blowing through cursors credits much faster then I used to, to the point I actually upgraded to the $60/month plan recently and still hit cap within days doing the same work I had been doing on the $20 plan just a month or two ago. Idk if they just grab more context now or what but the requests seem to have ballooned in token usage lately.
The 5hr refresh in antigravity on the other hand has worked surprisingly well, bouncing between claude and gemini I can easily get an entire day of working on various projects and still be good to go the next day.
The one minor downside I’ve seen so far has been the limited AI selection. Its pretty much just google and claude, but I was primarily using claude anyway in cursor so it hasn’t been a deal breaker for me personally.
I think I’m gonna cancel my cursor sub and move to Antigravity. At least until cursor figures itself out and stops the constant flow breaking updates. Opt-in to these changes needs to be the default.
