Gemini 4 Carbon is now available | Hacker News Reader

Gemini 4 Carbon is now available

blog.google

275 pointsmeetpateltech149 comments

Just checked my Gemini app, still only have 3.1 Pro, 3.6 Flash and 3.6 Thinking on Pro. Looks like it's Ultra only.

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.

I pay for Ultra and have been using Argon for a couple weeks. I just checked in both Agy and Gemini and Carbon is there for me now, too. I'll try it, but Argon is itself already so slow that I'm not sure Carbon will make sense for most activities.

On pro here and I can do about 2 design tasks a week before I hit usage limits with Argon. The usage limit with carbon is probably gone just loading the context.

Yep, I'm on Pro as well, and Argon been hugely useful, but I get about 10 prompts with it a week. And that means that I have to use 3.8 flash for most of what I'm doing, which is obviously not a great experience for any kind of long horizon tasks.

I've been using Argon (Medium) pretty heavily the past couple of weeks to create a website and an app and am not coming close to my Ultra limits. I've used High for a couple of code reviews and security audits.

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.

Argon is slow but surprisingly my progress is noticeably faster. If you aren’t prompting extra ambitiously with top tier models, you should probably just use deep seek.

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.

> Google's harnesses are so underwhelming compared to Claude Code,

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 ?

I wonder why my consumer Gemini app has Argon and 3.8 flash with a pro plan but so many people here don't have that.

Is the issue that consumers get it sooner or is the roll out random or locked to some other factor?

In one case I had to upgrade my Antigravity GUI to see new models, I guess because of an API change.

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.

I found out when I opened Gemini app on phone and it showed Carbon Medium as default

I have Carbon in my consumer Gemini too. I have a Google drive storage upgrade that I pay for that I assume is bundled with it - it's the only thing I can think of as I've certainly not gone out of my way to do anything special to get access (and so far, it is free for me to access it this way)

In true google fashion the situation is a total mess.

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.

My Google Workspace Gemini (with "AI Expanded Access") has 3.6 Flash, 3.8 Thinking, 4 Carbon.

My personal "AI Pro" Gemini has 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro, 4 Argon.

Pretty inconsistent.

This is the GoogleWay

considering how little you can use argon with pro, i guess carbon will be the same so why do you care?

To put a firm number on this, I went through 40% of my Argon quota with three small projects: update a cartogram with the daily election data from the California Secretary of State; read a bunch of campaign filing PDFs, property tax bills, and business registrations to create a graph of related individuals; and write an optimizer that snaps GTFS transit route shapes to where the roads actually are.

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.

Same. On PLUS subscription. One of those situations where you really want a human to talk to..

When's the last time you checked what is offered by the plus subscription? It's been a while now that it's the equivalent to ChatGPT Go (on price and offer), giving only increased limits but no difference in model access compared to free.

When I switched from the Pro to the Plus subscription, they did not state anything to that effect. I expected a smaller quota of their most recent flash models, and instead I got models that are so old they are useless to me, so I canceled the subscription.

It is the lack of transparency that gets me. With PAYG open-weight models, I know exactly what I get for my money.

> It is the lack of transparency that gets me.

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?"

I did check. Did you?

It literally says "Access to our Flash models" with no indication that they would see no updates for... How many months has it been?

I think you can understand why I was expecting the version of the Flash models to keep up with updates.

Edit: And back when I paid for my yearly subscription they also promised access to the Pro models. They simply took that feature away and offered no compensation. So, understandably, I am disappointed and disinterested in their offerings.

People see

"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 :/

They just took 3.1 pro away from me while I was asking about Cami Clark and Epstein (for due-dilligince), lol. Now I have access to 3.6 Flash and 3.5 flash-light. So I canceled my sub. Shame. I used to love using antigravity when the allowance was good.

I say the same thing on every thread. For 23e a month on ChatGPT I get access to the top models, almost always day one. I might not get a huge amount of usage but I get to use the flagship and try things out. On AI plus for the same price I get basically nothing, and models that are 6-9 months old. I don't understand why google is not using their size and leverage to get these models into consumers, with a useful harness and take market share.

It's not their business. Anthropic and OpenAI don't have millions of phones in the wild blasting the ever-loving hell out of Luna or Haiku to create timers and search contacts. They probably made Carbon just to help distill the next Gemini Flash. They are fine letting a _few_ people use it, maybe, but I bet they are also fine if no one does. Worst case for them is probably a bunch of $20 subs building things with their top model and taking compute away from Pixel 11 calendar searches or google.com "summaries".

This sounds an awful lot like Yahoo not needing to compete with Google on search because they have the most valuable "web property".

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.

Except Google already has marketshare for search and their Flash model does a decently good job there.

It’s 100% their business

They created the LLM!

It’s just bureaucracy, they’re not escaping involution

> On AI plus for the same price

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)

Is it location specific? I have AI Pro plan and I have had Gemini 3.8 Flash for months, and I got Argon about two weeks ago in Antigravity, and now I also have Argon in the Gemini app.

There is something to be said about the fact that Google tomorrow could release the greatest performing, most efficient, comprehensively capable LLM of all times and even if you are a paying customer of theirs, you could not access it across their interfaces. Not even going to mention the announcement to release delay, which given the numbers don't seem to have changed, I cannot comprehend why they'd do.

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.

I have Carbon in VS Code and the Antigravity CLI.

Seems like a completely fair and normal business practice to do a gradual roll-out first to trusted testers, then to Ultra, maybe to Pro later. Not sure why Plus and free users feel so aggrieved that they can't free-ride with those paying $200/month.

$200 a month is free riding on API users the $20 a month isn't free riding on the $200 plan.

Do you have any evidence that subscriptions are loss leaders?

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.

Paying 20$ a month for search already seems like a lot. That’s really what the 20$ tier is for

Hating on Google is cool, but pretending not to understand pricing tiers is weird as hell; when those 2 collide, you get this thread. Smart HNers, who I assume are fine understanding multidimensional tiers like the one Netflix has (price/resolution/simultaneous devices/ads) are seemingly very confused by Googles 2D AI tiers[1] (price/model availability)

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.

They are all available through OpenRouter for 2/10$.

Immensely more expensive than a sub.

I use Antigravity via Google Cloud and it's immediately available. I am not an Ultra tier subscriber.

It should be on Google AI Pro and Google AI Ultra subscriptions, I think.

https://gemini.google/subscriptions/

Checked my ultra and it’s there now. Too afraid to use it though since my quota would drop like a rock, maybe on Friday if I have enough percentage left over.

Update your Gemini app if its not up to date. You should have access to Argon on the Pro plan. Just keep in mind only Ultra has access to Argo/Carbon in Antigravity IDE.

Available through the API.

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 haven't tried Carbon yet, but Argon has been my favorite for asking questions or doing research on something I'm working on.

Out of curiosity, have you noticed being redirected to flash models recently?

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.

I haven't. I usually use it during the day US time zone. I get redirected to flash when I run out of usage on the Gemini web page, which has a separate usage meter from the agy weekly usage. I haven't run into it on the agy CLI that I use more often at night tho not a ton

A nice nod to simonw in one of the highlight videos;

https://www.youtube.com/watch?v=dNbkThtNzS8

That's awesome. Will models be pelican-maxed from now on?

Obviously. The next step after pelican-maxxing is AGI.

If this is not the end game when it comes to bike riding pelicans, I guess we are close enough. :-)

For OAI and Anthropic releases we usually get a "pelican" here in the comments pretty (suspiciously) quickly, which then also happens to be on top of the comments. I always found that pretty weird given that's not an actual model "evaluation". It's provocative and gets the people going, I guess.

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... ;-)

> Carbon made a pelican demo so you don't have to

https://x.com/sundarpichai/status/2120198811297694192

https://XNotMe.com/sundarpichai/status/2120198811297694192

It’s kind of bad at imagining what Mountain View looks like though ;)

The Waymo Ojai it bikes past is also adorable, great video.

Pelicans are solid, but every Gemini model for the past year has produced solid pelicans: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

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

It is such a massive miss from Google that it took them this long to release this model. I just ran it on my benchmark[1] and it matches Fable 5.2. They could have been first to get there, why didn't they?

That being said, it is still slower and more costly than Fable 5.2. So Anthropic wins in my eyes.

1 - https://bench.killswitch-lang.org/

That doesn't feel like a massive miss to me. No one will remember which was officially debuted first in a months time.

They'll care about the quality and the cost, etc.

It's about generating hype, not about remembering which was released first.

Don't see the incentive to generate hype. There's no IPO even if they convinced everyone they were leading AI it would be little benefit to them, in fact might hurt their cloud growth if people started jumping more into Gemini. They're better off keeping up for internal use and lying low publicly unless it gets very profitable

I apologize if this comes off as dismissive, but why would hype matter? Google is already everywhere, I'd imagine they don't really have to generate hype like a startup would.

Almost all devs already use claude or codex. Unless there's enough hype, they won't give gemini a chance.

I started using Gemini because Claude constantly refused. Codex is good but still refuses more often than Gemini.

I think the fact that Gemini is now in the same conversation as Fable 5.2 is a crazy feat for them. So I dont think this feels like a miss to Google. Especially knowing that they have their next version in the works.

> it is still slower and more costly than Fable 5.2

How is it more costly? It’s 4 times cheaper.

I have google AI plus because of drive storage and it’s ridiculous that the best model this gives access to is 3.6 flash, not even the latest flash model. Google either needs to stop this being branded anything “AI” or make better models available if this is what the entrance subscription gives

I don't understand why they do this

Why not give people a taste of the good stuff with a lower quota, then you may feel more inclined to upgrade.

I want to believe they want to slowly divert revenues to cloud by selling compute to frontier labs. Therefore they are intentionally reducing usage. But a simpler explanation is incompetence

People just complain incessantly about the low quota then

I got ai plus just before discovering gemini 4 is not included in the plan, of course it's not useful for me lol (I'm quite sure this was advertised, but not like, big characters saying you won't have access to the best models, just somewhere where it mixes with other things). Good thing I got it for free as a student

To be fair ChatGPT Go also doesn't give access to flagship models. But indeed staying at 3.6 Flash feels like big negligence.

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...

I was benchmarking 3.6-3.8 and it seems like they just tuned thinking levels to have higher token consumption. For example 3.6 medium was equivalent to 3.7 low. Your comment seems to support that

What did you think thinking levels did before you learned that?

I am saying that 3.6 to 3.7 there was nothing new except for moving the ladder by one step. Low was removed, medium became low, …

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.

> Getting started with Gemini 4 Carbon

> 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.

Argon was on Pro -- why not Carbon? Is it bigger and just priced the same?

Only on the Gemini app. This still remains the case for the Pro plan. Access to Argon under Antigravity was already limited to Ultra And will be the case for Carbon too.

Google really needs to get their shit together and sort out their product stack.

That's not the case since a couple weeks ago, I'm on a free Pro plan (bundled with my phone) and have accesss to Argon in Antigravity (but with different limits to 3.8 flash etc)

Is it 100$ a month for the minimum plan that has access

Yes.

> 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.

Yes, and it worked for a chat query or two. Then 5 hour wait.

I don’t think Google wants that 20$ tier market

I thought the 20$ tier was discontinued access to any model better than the free tier

No, what happened is free tier got kicked off 3.8 Flash.

Free only gets Flash Lite now.

The only major player that has regional pricing. I pay around $40. Free pro for students too. Comes with a whole lot of google one stuff on the side.

Oh good. I'm using Argon a lot since its release for development work on a unified CAD + CFD + DFM optimization software package for designing Group 1 to 3 UAVs. I've been using a mixture of Opus and Argon with Argon doing a lot of the coding work with Opus doing review/architecture. If this is actually good I could drop Opus.

With Plus subscription this is in the list of available models, but says: "Unavailable - Gemini 4 Carbon is currently available to Ultra plan subscribers."

Looks like it's the same price as Argon on openrouter, which is good to see. Probably was announced already somewhere that I missed.

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.

I wish they would either reduce the price of Argon, or provide a "weaker" model at a decent price.

At my work, Argon is deemed to be too expensive, and gemini-3.8 flash is just not good enough.

Deepseek is good enough for me that I use it for everything cost sensitive. I tend to switch to Google's models for world knowledge, planning, and complicated bugs that Deepseek gets stuck on. Argon is also a great writer in my experience, hopefully Carbon is the same. Seems like RL for code generation ruins most model's communication abilities. V4.1 Flash 1017 is a significant step back from the previous Deepseek model in that respect, despite being a noticeably better coder.

> Seems like RL for code generation ruins most model's communication abilities.

Just like human beings

Gemini series is easily the most slept-on/underappreciated model. I'm astounded some of the ridiculousness that some people put up with using Anthropic (Claudish, security problems, etc.) and OpenAi (nutso leadership).

Google brings the insanity of their product design and how nothing really fits together.

The extra training in SVG shows up in some of the web work I've done, in places where I want semi-scientific layouts (short of cad, but in the realm of design). I specifically used Argon due to it, and Carbon looks to be better due to the larger overall model.

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.

What’s the difference between /boost, /teamwork, and now /swarm?

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.

The documentation on swarm has a short comparison with the other options: https://antigravity.google/docs/swarm

I had some reasonable success running teamwork with argon orchestrator, 3.8 Flash implementor subagents, and follow up Argon/flash subagents running uno-reverse skill, mixed up with /goal

I needed something to switch the account where the usage is taken from, not the configuration.

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

Does it work with antigravity?

For some reason Gemini models are the laggiest for me, I use it for q/a and topic exploration, both through Gemini app and API , it often get stuck, usually in the middle of response(app) or at the very beginning before responding. I wish they would better improve on that.

OT:

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

Not sure how anyone is able to praise this model when so far we're seeing a p75 TTFT of 60s and a TPS of 10.

Model is completely unusable at the moment.

They say it's best for long running tasks, not for interactive usage.

Is this an intentional naming clash with https://en.wikipedia.org/wiki/Carbon_(programming_language)?

Maybe a stark reminder about growing data centre carbon emissions?

It wasn't enough for them to clash with Go!, now they're clashing internally?

https://en.wikipedia.org/wiki/Go!_(programming_language)

The page crashes my chrome

Same, Chrome crashes and Brave freezes. I don't have firefox on this machine for testing.

It uses more tokens than Argon? I already get very little out of my Pro subscription (using Antigravity) with Argon.

Yeah, I was barely able to complete a single project with Ultra 5x using Carbon...

Is gemini ever worth switching to from an anthropic plan? I feel like Anthropic is still pretty comfortably in front.

I use both, if you're heavily multimodal or need slightly stronger writing it can be worth it

Hm, interesting. What about the writing makes it stronger than claude? Also how's the usage limits

Claude just sounds like AI to me. Part of it is probably that claude is the baseline for LLM speak we all develop a radar for. Gemini has always been strong at writing (though I preferred early sonnet versions) and it seems that the stronger models get at coding, the worse they get at reasonable writing. They may fix that eventually when it's more focused on

Usage limits are similar to the competition

Hm, yeah, I can see that. I think it's pretty well documented that the more coding capability a model has, the more robotic it gets. I'll definitely check out Gemini for writing. Thanks for the insight!

I use both. I've been very impressed with Argon High on the $20 plan so I just upgraded to the $100 plan to try Carbon. On Argon high, on a 4 hour flight, I was able to build a pipeline to help us train a decision model off qwen3.5-9b and integrate the new flow into a rather complex codebase we have. I had previously tried with Opus 5.5 and was never happy enough with the output to productionalize it.

Oh wow. Which harness do you use for gemini? Also have you tried fable 5.2 to compare

I was using Antigravity at the time, which I don't often do. I've been using Fable 5.2 yesterday and this morning and am pretty impressed with it. Doesn't feel groundbreaking in anyway, just a small step up. They're seeming to be particularly generous with the usage right now, assuming since it's new. Haven't tried Carbon yet but will be today.

Yeah, I agree, I feel like fable 5.2 is more of an intermediate model before fable 5.5 or even the next opus model. Would love your input on Carbon once you try it. Also, I heard Antigravity has a bunch of issues; have you experienced any firsthand?

Just want to say I find Carbon waaay better than Argon. I keep noticing google model sumbissions on hackernews getting targetted by systematically negative comments.

We have an enterprise account and we can’t use this. What do you want us to say? Rollout has been incredible?

For coding or for which tasks?

The free personal Gemini account I use for things like tossing around NFL trade strategies, clearly got a lobotomy today.

It's like clockwork, every time a frontier lab launches a new high-end model, seems like all the models in their line-up take a hit as the cycles are reallocated.

> Using Google Antigravity’s /swarm feature, Carbon built an open-world 3D game where you explore Google’s Mountain View campus as a pelican riding a GBike

pelicanmaxxing is here.

> Using Google Antigravity’s /swarm feature, Carbon built an open-world 3D game where you explore Google’s Mountain View campus as a pelican riding a GBike — coordinating subagents to model 3D assets in Blender from Omni and Nano Banana references, build the browser game in Three.js with WebGL shaders, and pair procedural Web Audio sound effects with Lyria music.[1]

Amazing, can we now close this pelican benchmark and move on to different things?

Carbon and Argon are such similar names this is the first time I've registered they are different models.

Google branding strikes again

Compared to Sol/Luna/Terra or Opus 5.5 being better than Fable, the true test for AGI is a good consistent and clear naming scheme

That was faster than the previous announcement -> release cycles, right?

What's the recommended coding harness I should be using to test it out? (Recommended by folks here, not by Google...)

Surely it should be antigravity?

No, just like Argon it was released 14 days after the announcement.

Argon was announced 30 September, publicly available 13 October. Carbon announced 28 October, publicly available 10 November.

Antigravity is the obvious choice, unless someone manages to scrape out its prompts and API patterns then evals something better

Is anyone really surprised by Google (of all organizations...) definition of "available"?

No Google, keep it to yourself that way when you finally release it, you will be 5 models behind.

Interesting that on artificial analysis low thinking mode seem to cost more per task than medium.

yeah, see this very often. Less capable models tend to loop more and end up costing more per task, even though they cost less per token.

Are the limits low on the $20 plan because Google won’t make money?

Yes.

Well run businesses usually try to make money.

This has been available in AI studio via api for a while now.

Google does not know how to launch a model.

I see it in my Cursor Enterprise plan.

This is the problem. Google was ahead for a brief period where this model was unreleased. Since yesterday that is no longer the case. And what’s worse, it’s much easier to access fable vs carbon with the various plan offerings from the respective providers. I hope Google understands the urgency here and why it’s important to get their models out, if they want the accolades and user traction that comes with being at the frontier. I’d love for a much more competitive ai landscape and I’d love if Google could help make that happen.

Google makes it really hard to try their models. Guys, let me try it on pro or whatever sub. If I like it I will pay you more. I am already paying Anthropic 600 bucks a month and will pay more the moment they offer higher tier sub (I have 3x Ultra now) but I downgraded my Google sub because it's just so hard to try anything from them and it takes forever from the announcement to land in the app.

More Cartmanland marketing. It's the best place ever, and you can't come!

Still don't see it!

Gemini vis-a-vis Claude and GPT seems to have substantial unsolved hallucination problems. Hope it’s solved or it can’t be used for agentic coding.

People keep saying this, yet the benchmarks suggest it is actually the best at not hallucinating? https://artificialanalysis.ai/evaluations/omniscience

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.

The complaints about availability are at least somewhat true given the thoroughly brain-dead idea of announcing a model when it's released only to a very exclusive group - since it kills any positive rumours.

But I concur with the question about hallucinations. Worst I ever got since 3.7 flash if not earlier was model drifting too far when working on something very much new to itself and not being critical enough, something mk ostkt fixed by adversarial critical review subagent

Maybe they had insider knowledge that anthropic was about to release fable 5.2? The timing is a bit too convenient.

Maybe add (2026)?

Reply on news.ycombinator.com