They've been so slow to update their top models and roll them out that despite subscribing to the Gemini Pro plan, I use it so rarely that this is my last month. That and their product offering was so confusing I basically gave up.
AI plus is $4.99. AI Pro would be the equivalent at $19.99, and it gives you access to Argon (assuming 23e means €23)
ChatGPT and Google compete in the same market: a text box that answers your questions. The text box provider who can best answer the most difficult questions always wins because users don't care who they ask their easy questions to. Being a low-end text box provider is not a business model: providers either compete at the high-end or die.
They created the LLM!
It’s just bureaucracy, they’re not escaping involution
I'm not sure if I'm a sucker for paying $100/mo for this, and Google's harnesses are so underwhelming compared to Claude Code, but I'm not having a bad experience. Compared to the beginning of the year when Antigravity was almost unusable and Gemini Pro 3.1 was noticeably behind other options.
Could you please elaborate? A heavy user of Antigravity myself I had the opposite impression. Quite possibly I only scratched the surface of what Claude has to offer. For example how would you compare `/remote-control`? Would you say you prefer the Claude version ?
Not quite at the build me a whole app level but you need to definitely zoom out a few times and prompt from either a feature set or many overlapping details at once.
Is the issue that consumers get it sooner or is the roll out random or locked to some other factor?
At the same time, the CLI saw the new models without any change. Possibly that was just because my CLI install was newer than the GUI one. (Edit: oh, it's because my CLI autoupdates. Autoupdates for the GUI on Linux in an AppImage is... messy.)
Edit: But in this case:
> Gemini 4 Carbon is available via the API in Google AI Studio and Gemini for business for all paying developers and enterprises
I.e. Seems to be API only right now.
The gist of it is that there are many different "Pro" plans, and the model access/quota depends on things like whether or not you are a workspace "Pro" user, whether or not you are a discount/promo "Pro" user, whether or not you are a 3rd party "Pro" user, and then also beyond all that what country you are in. Common examples of these gimped "pro" plans are Pixel "pro" plans that came with pixel purchases, and Jio "Pro" plans which are discounted plans available through a telecom provider popular in India.
But if you have a normal Google account, and pay the full $20/mo, and live in a country that doesn't have tight regulations, you'll get the proper Pro account.
It is the lack of transparency that gets me. With PAYG open-weight models, I know exactly what I get for my money.
The Google One site lists all the tiers and their access levels. You didn't bother to check, and call the lack of a nag-screen a "lack of transparency?"
"Access to 3.1 pro" and they read it as "Access to Gemini Pro models".
Turns out when Google names a model in that subscription, they mean that model, not necessarily that class of model :/
1. Granted, the Google One pricing page is crowded with non-AI stuff, but for the AI model section is easy to understand what each subscription level gets you.
And by evidence, I mean something more than "at API rates my usage would have cost $1000".
Because Anthropic's SEC docs show that subscriptions are gross-margin positive.
Since Google has lower inference costs than Anthropic, I'd expect their subscriptions to be more profitable, not less.
What I will say is that their release strategy has led to this frontier model release post getting off the main page within 90 minutes, no other lab manages that.
Immensely more expensive than a sub.
I'm getting this quota for free from Google Fi so it seems like a square deal to me. But running out of quota mid-project can be irritating.
To add, it's actually surprisingly good. I have this unfinished paper draft that I keep feeding to various top AI's to see what it picks up on and what suggests to edit, and while Astra/Fable/Opus/Argon so far has been mediocre, Carbon's answer surprised me with some interesting points all the other AI's missed.
It's an anecdote, sample of one, but it got my attention.
I've not been able to get a _single_ response from Argon in over a week, no matter what or when I ask. I also usually am active later in the evening so figured this would be post peak activity anyway.
This happens not so much when Google or other labs release (not even on the Argon release post, afaik). Make of that what you will. Not sure if Google is sending a nod or a wink... ;-)
I also had it do this in Antigravity:
> Make a three js animation where yonkers stands up as a big mech (cool stand up motion) and stomps around manhattan and queens and Brooklyn, use openstreetmap mapping data
The result was pretty great: https://static.simonwillison.net/static/2026/yonkers-mech-st... (I had to prompt "build that into a standalone HTML file" to get that version").
Here's the same prompt (except I had a typo, I put "átomos" instead of "stomps", it figured out what I meant) against Fable 5.2: https://tools.simonwillison.net/yonkers-rises
That being said, it is still slower and more costly than Fable 5.2. So Anthropic wins in my eyes.
They'll care about the quality and the cost, etc.
Why not give people a taste of the good stuff with a lower quota, then you may feel more inclined to upgrade.
I wonder if they still offer cheaper non-AI Google One plans. I'd rebrand AI Plus as a storage plan with some free AI add-ons...
Initially I thought going from 3.6 to 3.7 will improve the model somehow while keeping the same speed/price characteristics but in fact it was the same if I just set 3.6 to medium - paying more to google.
> For developers: Build with the Gemini API paid tier in AI Studio and Android Studio, and if you’re a Google AI Ultra subscriber, you can try it in Google Antigravity and AI Studio.
> For enterprises: Try it in Gemini for business via the API or model drop down, Google Antigravity, or CodeMender.
> For everyday users: Try Carbon starting today in the Gemini app and AI Mode in Search if you’re a Google AI Ultra subscriber.
> For scientists & researchers: Apply for a chance to get access to both Argon and Carbon via Google Antigravity’s highest usage tier free of charge for 6 months (onboarded on a rolling basis), subject to eligibility, capacity, and selection.
> For cybersecurity defenders: Apply to join the Fairwind Program to get access to Carbon for cyber on a rolling basis.
Looks like the only way to access any Carbon tokens is to pay $99 / month, pay via API, or apply as a scientist.
Google really needs to get their shit together and sort out their product stack.
> It is also rolling out to all Google AI Ultra subscribers
However for what its worth the $20 plan got Argon like a day or so later.
Free only gets Flash Lite now.
Gemini is much worse than Claude, maybe I will give Carbon another try - just today I had to use Gemini because Claude went out of available tokens, so I copied some functions into Gemini to discuss and develop them with it:
- It mixes variable names in the output which I inserted correctly just 1 prompt before (like "NameOfficial" as in the class, became "OfficialName")
- It forgot things I posted in the prompt like some definitions in a file 20 min later after coming back from the phone, answer: "while I couldnt see Class XY, I made an assumption..." , heck NO: class XY was posted in the chat as well!)
- one time it inserted two arabic characters before "return XY;"
- It is allergic to C# interpolated strings if the construction is a little bit more complex etc.
Honestly - this is what Claude was in 2024
It's interesting that Opencode Go got Argon, I'm hoping they'll get Carbon as well. They claim that Argon is ZDR on their platform. That's surprising to me because openrouter doesn't have ZDR endpoints for Google's models any more, due to a change in policy.
At my work, Argon is deemed to be too expensive, and gemini-3.8 flash is just not good enough.
Just like human beings
This, 3d modeling and electrical/robot design are probably the three next frontiers of benchmarking and improvement. Medical and law are driving the boat due to the dollar signs, but engineering is the next most obvious hill climbing areas.
Anecdotally with Argon it would cut corners when attempting to one shot complicated prompts. Adversarial review would find hardcoded hacks and even bugs in physics code.
/teamwork did help with that but made it take EVEN LONGER
I’m hoping Carbon (with /swarm?) is going to help me with my physics and engineering. I don’t like having to cross check with Claude or Sol just to trust the results.
I made something simple & transparent named Switcheroo: https://github.com/FTCHD/switcheroo
Works with other AI CLIs like Gemini, and other various CLIs like Vercel (if you're managing multiple accounts like me), Netlify, Github etc
Model is completely unusable at the moment.
Usage limits are similar to the competition
pelicanmaxxing is here.
Amazing, can we now close this pelican benchmark and move on to different things?
Google branding strikes again
What's the recommended coding harness I should be using to test it out? (Recommended by folks here, not by Google...)
Argon was announced 30 September, publicly available 13 October. Carbon announced 28 October, publicly available 10 November.
Seems like as soon as there is any post about Gemini there are knee jerk reactions about how either you can't use it because it's not released (you can - it is released), you can't use it because you can't buy it (you can - consumer plans or API or just even OpenRouter), or that it hallucinates worse than other models (not according to https://artificialanalysis.ai/evaluations/omniscience that actually says it is SOTA in this regard).
I've used 3.8 flash a lot as my daily driver (...eagerly awaiting flash 4 or whatever it ends up getting called Zirconium or Caesium or something other nonsense I am sure) and I can't say I've noticed any significant hallucination issues that everyone seems to think are so prevalent.