I don't mind Carbon, Argon kind of throws me off because my brain autocompletes to Argonian... I still have no idea what the different ones from OpenAI mean. Can we just go back to "Silver, Gold, Platinum" for naming things? Something simple we've all understood, always.
Other sources say that what was publicly released as Argon was already the second version named Barium internally. So there was a separate internal Argon and public Argon was Barium. Not confusing at all for them!
https://www.reddit.com/r/GeminiAI/comments/1x292ly/a_new_che...
Still better than Anthropic. If I create someday doomsday device, its turn off code will be to sort Anthropic models by strength given only their names, without access to the internet.
Serendipitous timing, but I've been building a little chrome extension [0] for doing exactly this sort of thing (highlighting tables automatically, especially when numbers don't necessarily correlate good=green, bad=red). Model pricing and comparisons are one of the main things I checked (low cost good, high benchmark score good).
yeah, they've been quite logged-out hostile for a few years now. not quite as bad as instagram, but they seem to be holding that as their guiding-north-star...
(Edit: For some reason the previous jpeg link: https://files.catbox.moe/yjqa0a.jpeg wasn't working and catbox.moe was just downloading 4kb white screen so I converted the image to webp and reuploaded it on catbox)
Without color blindness it looks fine, green-yellow-orange-red. Having checked on a random website what a protan colorblind person might see, yeah, I get your confusion...
"The Fairwind Program gives high-priority defenders (like governments, healthcare providers, and telecommunications services) early access to advanced models that help them build better defenses, before new threats arrive."
Again, no access to open source developers. The AI companies want bugs for dramatic public relations! Also, of course there aren't enough bugs so they have to be rationed for the "defenders" and future models.
Given how quickly the field is moving, showing benchmarks without releasing the model is quite meaningless. If they decide to release it in one month, there will probably be a stronger competition out there. And the benchmarks won't look as good.
I can't tell if they're withholding access to manufacture some sense of scarcity but the delays are really prolonged. I'm nervous to use Gemini models anyway, as its attached to my Google account and I've seen too many people having their whole account banned for dubious reasons (with little recourse).
Also the whole "we're only releasing it to cyber defenders first" framing is weird, is any Gemini model an actual cyber risk?
They released Argon shortly after announcing it, so I think some benefit of the doubt is reasonable here. They claim they're already evaluating it via the government.
I suspect they're announcing it for two reasons: to get ahead of all the leaks; and because it's earnings report day.
I think the majority of traffic is from Enterprise customerd that goes through their API. Their consumer facing products do lag, but I don't think they care.
Just remembered i haven't tried argon yet, they announced argon then and delayed releasing it, didn't even know they've made it publically available. I don't understand why google thinks this is an awesome strategy in a market where there are other really good models.
For what it's worth: Argon is a pretty good model, and this is coming from someone who just sticks with Opus 5.5 mostly across the board. For the current introductory price, it provides an exceptional value. Looks like they're trying to release Carbon (which is apparently a significant improvement) at the same price as Argon, which would be a crazy value.
It's worth trying if you have easy access to it (we use Cursor at work, so it was just a matter of picking a different model for me). Don't know if I'd go out of my way to get a Google AI subscription for it, but if Carbon ends up meeting these expectations, then it may be worth doing so.
It's best to start with a relatively clean GEMINI.md too. Many configs optimized for Claude doesn't work well because Argon didn't have those same failure modes.
We're working on a battle system for local coding agents [1] and Argon is really bad. Literally the worst model in our initial ranking, and Deepseek Flash, Muse Wonder, Grok are all in there.
You're right Arena.ai has far more data. Our data is actual local coding agent head-to-head battles, so I think our quality is far higher.
Their agents leaderboard is literally just randomizing the agent and looking at metadata like tool call errors. In my opinion, it's poor quality data in high volume.
Their web dev leaderboard is in cloud containers, most people don't run agents like that.
I ran a statistical test on our data. Deepseek vs. Argon. Posterior probability DeepSeek 4.1 Flash is stronger: 99.1%.
Well, my perspective is somewhat narrow (I use it primarily for software engineering and pretty technical data engineering work), so it's possible that Argon is well-suited for the work I do but not for other tasks. I suppose it's just more data points to consider.
Our data is all software engineering run via local coding agents, but yeah your specific use case could be one where Argon is stronger. (Or your personal preference.)
I paid for their lowest-cost subscription, thinking that it would give me a small quota of use of their most recent flash models. Instead, they give me an unnecessarily large quota of their old models, which I have precisely zero use for given that open-weight models are better and cost next to nothing.
Consequently, I have zero intention of renewing my Gemini subscription. What a dumb business choice they made.
I have a pro subscription and got Argon about two weeks after the announcement, I'm on the annual plan, got the 30% discount by switching it from monthly.
Usage on antigravity is not included, usage via chat is absymally low.
It's a good model though.
I have a yearly plus subscription and I am still seeing 3.6. I am not renewing it once it expires. Why not give subscribers a small usage quota of their latest flash models?
> Reducing shortcuts and loophole exploitation: With Carbon, we further mitigated shortcut attempts in training while ensuring high-quality coding performance. Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4% — all while improving overall accuracy.
This is probably the single most important thing to make frontier models usable right now, so I'm happy there seems to be some progress. This was the previous Gemini's biggest problem IMO
It's the speed of these gemini models that I'm most impressed with. I'm finding speed to actually be a larger limitation than intelligence these days of running multiple agents 24/7 building stuff.
>Resisting prompt injection: While Argon previously led the Gray Swan’s Indirect Prompt Injection (IPI) Benchmark in prompt injection resistance with a very low 0.7% attack success rate, Carbon achieves 0.0% — resisting every evaluated attack attempt.
"We patched every vulnerability we knew". Meanwhile, prompt-injecting Argon (or any other model) is still trivial, and Carbon won't be an exception. Because the attacker is not limited to things Google has evaluated.
AI shops are blind to anything that's not a metric.
I'm surprised no one at Google realized that both OpenAI and Anthropic's naming schemes follow a fairly clear progression and that they should do something similar to reduce confusion. I would have assumed the opposite given that Argon has a higher periodic number
Both seem to have the same pricing: $2 in / $10 out initially, rising to $4 in / $20 out after the introductory period. That made me wonder whether Carbon might be Argon with additional RL training. What’s the Opus/Sonnet comparison based on?
I don't think that's right. Carbon and Argon have the same price.
I suspect they're using a new naming convention because they had previous contractual obligations tied to the "pro" and "flash" naming, and by switching to a new naming schema they avoid those commitments.
Other comments on this already, so this is more of a confirmation.
I'm not going to comment on others use of Carbon except to say some article that came out before Argon's external launch saying Googlers hate it was nonsense. Not to say everyone loved it all the time.
I use Carbon heavily and am quite happy with it. For complex designs I may have it and Opus iterate back and forth on the design. They have their own strengths and Opus 5.5 is great, but I use Gemini far more than I do Opus at this point. I know a lot of the jagged edges of Gemini better than Opus though, so take that as you will.
People I work with either use the dogfood versions of Pro models, the latest public Gemini versions, or Opus 5.5. It's a personal preference thing, we have public usage dashboards and you can see some people are 100% Gemini, some are 100% Claude, some are using both.
Google offers Anthropic models hosted on GCP through their Agent Platform (fka Vertex AI) - presumably any internal Anthropic use never crosses the Google boundary.
I personally use 3.8 Flash most of the time and Argon/Carbon when I need to do deep analysis or research an ambiguous problem. Opus/Fable is used when I need a second/third opinion.
From what I heard from a Googler, Claude models are still the most popular for coding internally but it’s a couple of month old info so take it with a grain of salt.
> Building on Gemini 4 Argon, which was launched two weeks ago, Carbon illustrates the pace at which we’re pushing the frontier forward. Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem.
Is it a variant of Argon? A different model? The next version? It seems like this is a separate class (like Anthropic's Mythos).
They seem to go with Android-style alphabetic order code names instead of pointed numbers (4.1, 4.2...) and drop "B" for some reason (which is creating this confusion.)
Arrg but the period table already has a natural order, sorting elements alphabetical isn't exactly intuitive - although fitting given the parent company.
We all need to go back to numbered versioning. I've been on macOS and have to check every time I need to know what OS I'm running. High Sierra, Monrovia, Yosemite, who the fuck knows I've spent two days in California my entire life. I also shouldn't need to know what minor version I'm on to discuss my OS's capabilities. 26.2 is what I'm on now.
I applaud that Google is making state-of-the-art models, and no doubt they want credit for that. But it also highlights how hard it will be to get back the develop-class. Who is going to migrate from Claude's for this without an enormous incentive? Claude has the trust and the ecosystem for those who want the very best. Codex is clipping at its heals, and as arguably a better total value.
Love the options but also I hope Google understands that they need to now compete on many dimensions for developers. Google probably will still win huge shares of the consumer market, though frontier intelligence will have little to do with it, other than the RSI aspect.
> Who is going to migrate from Claude's for this without an enormous incentive? Claude has the trust and the ecosystem for those who want the very best.
Both A and OAI regularly screw devs on quota.
Switching for many is easy, you just run agy cli instead of claude/codex, and enjoy Ads free youtube and music + 20TB of google storage.
I had to google how the hell you even use these. Apparently “antigravity” is their suite of agentic products and Gemini CLI is dead and it’s the same situation where a google AI plus subscription is only useful inside the antigravity tools. No Pi/OpenCode support at all.
My view is that the llms aren't sticky. I switch from one to the other daily. frequently auto which selects the cheapest option. When codex came with sol the entire team switched instantly. Then opus 5.5, the entire team switched instantly. We have a channel on slack where somebody says which is the best cheapest model today and I usually use that.
> Gemini 4 Carbon will launch at the same introductory price 1 as Argon at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.
So more capable and same price? I can at least appreciate that
While there are a hundred different evals it seems to me that we've settled on crowning the #1 coding model as the best model. So in those terms this is still behind Opus 5.5 and Astra.
> "Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59."
Basically everyone[0] thought yesterday's update looked suspect, even before we knew about Carbon.
Did Google coordinate with AAI on any aspect of this in a non-public way? Is this the sort of thing a publicly-traded company would be obliged to disclose, if they were involved in it? AAI isn't a regulated company (private), but the stuff they print influences stock prices.
Dont companies always talk in private? Like, Apple having a business meeting with Meta is not nefarious nor out of the ordinary. Companies coordinate on things all the time. Why is that surprising to you?
That market resolves using the arena.ai text leaderboard, where Argon was already #1 over Opus 5.5. Coding is way more economically consequential, and I think the current gap between Anthropic and Google is very wide there (at least, before this Carbon release)
That's a good point. The coding specific question (which still uses the arena.ai leaderboard, just the coding version instead of text version), has Anthropic at 66% at Anthropic/OAI at just shy of 80%. That's probably more representative even if Arena.ai isn't my favorite benchmarking site.
Actually....looking in more depth I may have to upgrade to something stronger than "not my favorite benchmarking site". The first OAI model on there is GPT 5.4 at 31....that's not really credible.
There aren't any _good_ elements that start with the letters D or E. Imagine if they name the next one "Dysprosium" where the Greek root translates to "hard to get at" or "difficult to access" reflecting how difficult it was for its discovers to isolate it in the late 19th century, but also sounds kind of untrustworthy to give an AI model such a name.
I can understand alphabetical ordering in some circumstances, but if the point is to indicate which model family is lighter or heavier, surely atomic numbers should be taken into account?!?
Argon - 18
Boron - 5
Carbon - 6
I don't think I'll be able to cope with this level of elemental disorder.
Note that the benchmarks seem to be cherrypicked. For Argon they were also cherrypicked, but differently, when Argon released. For the new benchmarks (cherrypicked for Carbon), Argon is now behind GPT-6 Astra in most of these benchmarks, even though Argon was ahead of Astra in the benchmarks cherrypicked for Argon. So apparently they couldn't find benchmarks in which both Carbon and Argon was overall ahead of Astra.
I'm assuming this is a post-train of Argon ("Building on Gemini 4 Argon, which was launched two weeks ago, Carbon illustrates the pace at which we’re pushing the frontier forward."), but does anyone know for sure? The post isn't clear.
tbh i might be one of the few who actually likes gemini since 3. it wasnt as good at coding compared to other two but it always wrote the best, and talked the best, providing very good conversation experience. it works incredibly well in my customer service agent harness and my users have been very satisfied. Now Gemini 4 is incredible, but the same old issue: a powerful model, in the worst official harness
deepmind should get rid of those antigravity ppl they're a joke and sabotaging the model's reputation.
Will this make it to OpenRouter? Running your own benchmarks against these models is surprisingly frustrating when you don’t want to create hundreds of accounts for each AI lab’s API.
I wonder, with all these releases and with all impressive open models, is Mythos even relevant anymore? Did it even exist or was just a, ahem, "run 100 concurrent requests and select best response"?
I’m itching for more smart cheap models. I have a mid level Claude and ChatGPT subscription and I’m blasting through my usage, and I’m too cheap to pay for the higher tier.
I use DeepSeek too but it’s more of a workhorse model, requires a lot more guiding and sometimes struggles to get where I want it.
There are so many benchmarks now, it’s hard to know which ones to look for. Google has listed a bunch that they’re winning at, but idk if those are cherry-picked or not.
Google is way more risk adverse than the smaller labs, because the punishment for slip ups is dramatic.
They cut down 1.5sq miles of man made logging forest that was rezoned to industrial land (and sold to google), without getting the right permit first, and it's been international news for almost a month. Anthropic and OAI are still small fish, at least publicly, so they can get away with far more. If Gemini was doing the shit OAI was doing with agent hacks and the like, there would be congressional hearings and $10B in european fines.
When you are a megacorp, perhaps here even the megacorp, every mistake you make is worth millions to someone, the government, the media, individuals, or all of them
It's totally possible, especially with meeting all the major heads had with Trump a few weeks ago.
I imagine they all want to avoid uninformed knee-jerk legislation, so preventing a "three mile island" type incident that would freak out society and congress would be common ground alignment between all the labs.
The air around AI is already dense, but it would suffocate the industry overnight if something like a large scale financial attack or grid attack took place.
> When you are a megacorp, perhaps here even the megacorp, every mistake you make is worth millions to someone, the government, the media, individuals, or all of them
I don't buy it.
SpaceX is not exactly a small corporation, and they nevertheless strike me as far less risk-averse than GOOG and they deal with actual hardware that actually effing explodes.
I'm going to say that deforesting without a permit is quite a bit worse than not releasing vaporware.
This isn't about a "slip up", it's the modus operandi for Google since it was young. Press releases instead of real releases. I can't count the number of times I've clicked on a "launch" story from Google on HN only to end up on a blog post with an embedded youtube video, and nothing else. Google just doesn't get it, and this is one of the many ways that manifests.
The point is that if the local commercial developer cleared the forest without the permit, it probably would not have even caught the ear of the local news. They pay the fine and move on. Happens all the time in bureaucratic systems.
Clearing the forest and failing to get the permit are two distinct events here. Google is in trouble for failing to get the permit. Not for clearing the trees. Everyone on both sides was aware the trees were coming down. Think of it like camping at a campsite without getting a permit first.
Failure to get a permit is non-news for almost every entity on Earth. Except when you are megacorp, hence my comment.
I think Artificialanalysis should have a policy of only including available models or something. They already include Gemini 4 Carbon now even though nobody can access it. Google can celebrate that they have the highest scoring model. Fable 5.5 and Astra 6.1 will probably be released before Gemini 4 Carbon and might well outperform it...
Deepmind creates these, Google Cloud productizes them.
If Google Cloud was in charge, Gemini 5 would be public, and they wouldn't be messing with Gemini 4 nor 1M context when they have 10M internal. C'est la vie.
A 10M window would require dozens of GPUs (or TPUs) dedicated to each query. That is something that Google could afford internally but wouldn't be able to scale to an external product.
Everything at Google rolls out incrementally to different populations of users so no matter what feature they are announcing there is a good chance you can't see it yet. When they announced Gemini 4 Argon it was 2 days before I could use it.
Ha, I didn't see your comment before I posted mine [0], there were still only 2-3 comments when I opened the thread but yeah. I completely agree. I don't think they are lying, but I still don't love the "it's the best, but you can't try it".
Argon 4 is really good in my opinion, especially in chat. Usage is limited, but one big difference is that it seems to better understand what is and isn't worth mentioning.
It has better tuning towards not adding unhelpful caveats and irrelevant details. When I told it I was unhappy with some slop being dumped into some repo markdown files from Astra and Sol it actually went ahead and tastefully cleaned them up in line with how I expected them to be. Reductions are normally so bad from agents.
One big thing is that it really feels like talking to someone smarter than me. Other agents just seem extremely knowledgeable but Argon felt smart.
Google really tried to make it sound like they're not in the AI race, but suddenly are pushing out huge challengers in the space. I originally took them to be fully focused on the consumer side, but it looks like they're going after it all, now.
They really were the sleeping giant and I think everyone understood that - OpenAI especially, hence their need to scale up hardware as much as physically possible. Google is trying their best to keep their heads down and focus on their lane. It's starting to pay off, but I'm sure they've felt the heat more than once.
I don't trust AA scores after the last update to the index. Specially now that they are including private benchmarks that no one can test by themselves but only trust the entity doing them.
> Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index [0] at 59.
If you are going to hyperlink, maybe make sure that your model is actually on the page.
Also, the screen reader has a bug ([[duration]] minutes instead of the actual value) and you have once again announced a model we cannot use.
Edit: Both have been fixed after posting this. Maybe a Gemini 4 Carbon Agent has read and fixed the screen reader bug xD
I cannot fathom what motivates a lab, whose releases for the last year haven't been competitive, to shoot themselves in the foot like this. All this does is make it easier for competitors to snipe your actual launch, even if there was potential with this model (doubtful given 4 Argon, 3.1 Pro and 3 Pro), now everyone knows what exactly they need to release eval results wise to beat Carbon. There is this old idea of "Your competitor always is the first to have an idea what your next product will do", but in this case, Anthropic and co don't even need to guess, they got specific numbers to target.
Announce your models the moment they are available on API and in the subscription for all paying subscription users (cough Argon cough), stop with the cryptic tweets to drum up hype you know full well you won't live up to and make models that tool call, task adhere, task infer and handle long context tasks well beyond simple back-and-forth, chat style prompting. Reminder, during Ox Alpha, actual Google employees implied that was their model. How is this something anyone would believe is a good idea and how are they still allowed to tweet anything after? Please, stop that, it isn't helpful.
If frontier isn't your thing, stop wasting money on moonshot pre-train runs, fund research and open weight via Gemma and sell your inference hardware to other labs, that's far more lucrative anyway, as seen with Colossus 1.
Oh come now, this rolled release is now the standard for frontier models. Also, the fact that AA hasn't updated their page is not a scandal, just a launch day hiccup.
> Oh come now, this rolled release is now the standard for frontier models.
Sort of? Yes, OpenAI has been bad about slow rolling but Anthropic's last few models have been available essentially at the same time as they announced it. In most cases I updated Claude Code and was using the new model right away.
Compare that to OpenAI who sometimes says "Available to all Plus/Pro/etc plans" without saying anything about "rolling out" and then it's a crap-shoot on when access will actually roll out. That said, they have gotten better, it used to be 1-3 weeks sometimes to get a new model/feature they announced was available for your plan.
And then there is Google... Who loves to talk about models like this and then give no timeline for release. Even when they do release the models it's miserable to find out how to even use it. Is it in the API? Which API? Google Cloud/Vertex/Gemini Cloud? Whatever it's called this week? Coding plan?
No, what Google does (or rather fails to do) in its rollouts does not compare to OpenAI/Anthropic.
As for the AA page, it's just embarrassing. Any smaller company doing that would get roasted, Google doesn't get a pass, learn how to coordinate or wait for something to actually be true/live before posting.
There was none for Fable 5.1 and the time between Astra announcement and availability was a few days.
Both OpenAI and Anthropic also have a bit more credibility to make such moves, making minor delays while unnecessary and self-defeating in my view, at least not as egregious.
I will say that I dislike how OpenAI releases model only for Work and Codex but not via Chat, Anthropics "everywhere all at once" strategy is I feel the most paying user friendly.
Gemini 4 Argon took weeks to come out and when it did, paying Google AI plan users couldn't access Argon on Gemini or Antigravity.
I imagine there is immense pressure to announce taking the #1 spot while they still have it. It's probably better than releasing a model that hasn't been properly vetted, but I agree it's annoying.
Fair point, though if they are still vetting, that'd likely also entail some post training or at least the addition of a filter, should they find something they do not like. As we've seen with Fable 5 and Opus 4.8 Fallback, that can significantly affect user experience and benchmark scores.
So if they are still vetting and changes can thus happen, that'd mean the score could change for better or worse. Not sure they'd want the embarrassment of a #1 score changing after announcement.
AA has also lost all credibility to me with all of their index changes being unannounced and following an open weight model getting close to the top spot. I gave them the benefit of the doubt for v4.2, but to do this again immediately after Kimi K3.1 got into the top 5 is a little too suspicious for me. For Opus and Astra to stay at about the same score while everything else drops feels like they are adjusting weights specifically in their favor.
Even in the optimistic case that this isn't some conspiracy, it highlights how arbitrary their scores are. I had trusted their previous ratings and their new index shows that even AA doesn't agree with them. What's to say this new index is any better and won't just change drastically again?
gemini 4 carbon has now come on the artificial analysis page. I think that I had checked it 5 minutes ago and it wasn't there.
For historians, The blog post is 31 minutes ago so we can say that it was true for 20-30 minutes.
Slightly embarrasing but it seems to be the case that the model is actually worth what they were saying on AA. It's just that it is not available to people and I think that Anthropic and OAI will release a better model to public before google might.
AA has tons of its flaws too and the recent shift to their new v5 has been nothing but controversial for many.
The fact that frontier models actively attempt to subvert evaluation harnesses (0.4% attempt rate still remaining) remains fascinating and slightly unsettling.
I wonder if the google ai team has to do some RF to scrub the whole "Don't be evil" thing from its all-written-human-knowledge weights before they can get it to attack.
When I mention a preference for brevity in a comment offhand and it snowballs into a rule 8 sessions down the road in a PR review that requests the removal of 90% of a colleague's docstrings.
I can't imagine all the "push back" on the "load bearing" Code of Conduct from their corporate foundation documents.
Makes sense - Flash models are (mostly) distills of the frontier model capability.
If they were able to make flash 3.8 so good off of their 3.1 pro model, flash 4 will probably be really good. But ofc you'd want your pro model to be as strong as possible in coding so that flash can take all those gains and efficient-ize them.
Oh, look! Another model that's supposedly amazing at Cybersecurity.
Too bad you won't be able to use it to secure your open-source projects. If you try it'll throw up a prompt telling you to use some initiative you can't use or afford while the provider collects the findings.
What a racket.
> Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59.
Clicks link
Carbon isn't anywhere on the page and you can't toggle it on either. I'm sure it's just timing and it will be there soon but still just look amateur, which is about how I feel about Google and AI overall. I know the models have improved but I still have a bad taste in my mouth from testing Gemini in the past and I still feel that Google has an overly complicated "how to use our models" that changes way too often. That coupled with "you could lose access to your google account if we don't like what you do with Gemini" means I have little desire to test it out.
On top of all of that I'm completely over announcing a new model that we can't actually play with or try out. I don't think Google is lying about how well their model scored exactly but it still has big "my girlfriend goes to another school, you don't know her"-energy.
Seems great? Is Google back in the ring? Now they "just" need a competent harness...
But damn, is (talking about) releasing a new SOTA model two weeks after the previous one stupid. Do they not plan these things? Argon shouldn't be released at all if you're going to put out "argon 1.1" few weeks later.
Can’t stand these paternalistic companies gatekeeping new models because “cyber”. We saw how this went with the PHP vuln last week with defenders using special / early-access models and then saying “hey there’s a vuln we won’t say what it is but it’s there and we won’t patch it until tomorrow”, and with the huggingface devs having to use Chinese models to defend from OpenAI because their American models wouldn’t help because cyber.
Any bad actor worth their salt is already a) going to have the weights downloaded, forever, to various models that are good at hacking and b) is out of reach of the U.S. government anyways. A ban on open weight models, plus paternalistic gatekeepers of cyber models like the big labs, just opens up Americans and American companies and municipalities to hacking from abroad. Or I guess to a really nice racket where only the big labs can offer a “we’ll continuously monitor and improve your cybersecurity for millions of dollars a year”.
“We're taking a deliberate approach to rollout and are actively engaged in the U.S. government’s voluntary process for pre-release model testing. Starting this week, we will make Carbon available to a subset of cyber defenders and will gradually phase access to additional trusted partners through our Fairwind Program. We’re continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible.”
I think you need to either be naive, selfish, or ignorant to think that giving the equivalent of cybersecurity nuclear weapons to everyone is smart plan.
Obviously we want to be vetting the parties that get to use these weapons, because we want them to only be used as tools.
Mind you that Kimi in it's latest release nerfed cyber capabilities, as they stated. It doesn't make sense for China to be loose with this either.
And these models are deployed in a large number of data centers, with thousands of employees and on millions of GPU:s. The GPUs are built abroad and shipped to the datacenter.
Billions are likely spent by China and Russia to get their hands on these weights, one could argue that ripping off the band-aid now is the only way forward.
My general impression of all Google models I've tried so far is a general "goody two shoes" feeling which I find very, very frustrating.
I'm going to experiment with this one (assuming they ever get around to actually releasing it), and I'm really hoping the lame cultural aftertaste is gone.
Great to see Google reclaim the frontier, however, as usual with most Google launches, an announcement for a product that no one can actually try outside of a tiny circle of users.
This feels like a pretty big deal. Quite a few people sleeping on argon because it doesn't come via a subscription, but it's noticeably better than pretty much everything in my own weird game based benchmark (https://botbowlbench.com/)
It's clear that all the labs have stagnated and have no idea where to go from here. All their results are basically equal now, they're benchmaxed to fuck and there are zero new architectures.
Is this the best and the biggest model from Google now? I have a hard time understanding this stuff with all the various tiers and “best” terminology use within tiers.
The problem with Argon, that makes it feel like older models to me, is that it stops and waits for confirmations far more often. If run (in a VM or container) in the dangerously skip permissions mode, it gets a lot less annoying. It feels quite smart and thorough, to me, though still lagging Opus 5.5. I'd be fine using it instead of Opus for most things.
This post explicitly says that's been improved in Carbon. So, here's hoping. Google currently has the most generous usage limits on their $20 plan among US providers, so it's a pretty good deal.
> Google currently has the most generous usage limits on their $20 plan among US providers, so it's a pretty good deal.
Are you saying when you use Argon in antigravity it has generous token usage? I forget if I tried it with antigravity but my experience with using it as a chat bot was the usage limit was terrible.
Yes. For similar tasks using Antigravity+Argon vs. Codex+Astra vs. Claude+Opus, Gemini has the most generous limits on the $20 plan. Since the ChatGPT limits got nerfed last month (or whenever, I can't keep up), the $20 ChatGPT plan is very stingy. With similar tasks I'd see 2x-3x faster consumption using Codex+Astra vs. Antigravity+Argon. I could casually code all day on the $20 Gemini plan, using Argon on Medium, without ever bumping into the limit, which is not true of ChatGPT or Claude $20 plans (but used to be true of ChatGPT). If I were using automated tools and doing a bunch of things at once, I'm sure I could saturate it, but I don't think most humans are very good multitaskers even with agents doing most of the work, so I tend to stay pretty engaged with what's happening and while the agent is working, I'm spending time looking over the previous work, testing, editing docs, etc. I'm not juggling five projects at once, most of the time.
The order of generosity, as it seems to be from my limited testing is currently Gemini->Claude->ChatGPT. (And, if we're including Chinese subscriptions, Kimi is below all of those. Very stingy.)
The Google naming convention is so nonsensical it's comical.
With Anthropic and OpenAI, it's obvious immediately which are the better models. I.e. Sol is better than Luna, or Mythos is better than Fable.
But Google, let's pick Argon, which to most folks means mostly nothing. And then... Carbon! So which is better, Argon or Carbon? Nobody fucking knows. Plus neither sounds cool or viral or mysterious or even provokes curiosity.
When Fable was first leaked it sounded different and intriguing... Carbon just sounds... boring.
> it's obvious immediately which are the better models. I.e. Sol is better than Luna, or Mythos is better than Fable.
It is not obvious to me at all. I like Apple's naming more: iPhone Pro Max is better than iPhone Pro which is better than iPhone.
Sol vs Luna, how am I supposed to know which is better? Is sol a word, what does it mean? Same with mythos and fable. How am I supposed to keep track of these names across multiple companies? Why can't they use Claude vs. Claude Pro vs. Claude Pro Max?
Someone in some previous thread tried to summarize, e.g. "Sol < Astra, except when 6.1 Sol which is more that current Astra, but Terra 6.1 > Sol latest, and..."
So no, OAI scheme is also confusing as hell. And they are going to add even more confusion when they release presumably Nova.
Tho I think it’s definitely not obvious at first glance. And requires some thinking to order if you are even familiar with the terms in the first place (I had to look up opus)
So now we are just showing some tables with each model?! Who even used argon in normal day to day stuff? From the time they opened Argon publicly to now, it’s like: three weeks maybe?! Most these benchmarks weren’t realistic to start with, and now it is just meaningless tables, not to mention the model degrade from the release date to after few weeks.
But I come from a latin language background, so maybe I'm a bit biased there.
Size wise Astra is not categorically bigger than Sol, both in absolute terms and the observed size.
Maybe it thought the names are clear and intuitive (for an LLM perhaps).
https://i.imgur.com/IDFGuJU.jpeg
[0] Practical Data (https://practicaldata.dev/ / https://chromewebstore.google.com/detail/practical-data/oeoo...)
https://files.catbox.moe/2aqfk7.webp
(Edit: For some reason the previous jpeg link: https://files.catbox.moe/yjqa0a.jpeg wasn't working and catbox.moe was just downloading 4kb white screen so I converted the image to webp and reuploaded it on catbox)
https://imgbb.com/
https://catbox.moe/
Anyway, Imgur creates so many problems that it should be retired.
On the other hand I'm a strong protan colorblind guy (red doesn't register to me very well at all).
https://file.garden/auI5UdLIWwZSu5X8/pilestone.png - result of said simulation.
Again, no access to open source developers. The AI companies want bugs for dramatic public relations! Also, of course there aren't enough bugs so they have to be rationed for the "defenders" and future models.
Also the whole "we're only releasing it to cyber defenders first" framing is weird, is any Gemini model an actual cyber risk?
Just because you saw your friend move to a new country for a job doesnt mean you will make the same choices, even if you are similarly qualified.
I suspect they're announcing it for two reasons: to get ahead of all the leaks; and because it's earnings report day.
Especially on their Plus tier for Gemini.
I think it's a fair concern that this might take weeks and trail Fable 5.5.
It's worth trying if you have easy access to it (we use Cursor at work, so it was just a matter of picking a different model for me). Don't know if I'd go out of my way to get a Google AI subscription for it, but if Carbon ends up meeting these expectations, then it may be worth doing so.
It's best to start with a relatively clean GEMINI.md too. Many configs optimized for Claude doesn't work well because Argon didn't have those same failure modes.
Top is Opus 5.5, followed by Fable.
[1] https://github.com/bottomless/agent-duel
https://arena.ai/leaderboard
Their agents leaderboard is literally just randomizing the agent and looking at metadata like tool call errors. In my opinion, it's poor quality data in high volume.
Their web dev leaderboard is in cloud containers, most people don't run agents like that.
I ran a statistical test on our data. Deepseek vs. Argon. Posterior probability DeepSeek 4.1 Flash is stronger: 99.1%.
For me it also works the opposite way, when it actually comes out eventually it’ll already be old news.
Consequently, I have zero intention of renewing my Gemini subscription. What a dumb business choice they made.
for whom?
Doing this two hours before today's earnings report.
It's probably going to be a terrible earnings call in terms of capital expenditure, cash flow and new borrowing.
Now you know why they released it just before the earnings call.... :-D
It has a seperate weekly limit that is very low. Maybe it can be pulled off that usage tier and just regular usage since Carbon is being released?
This is probably the single most important thing to make frontier models usable right now, so I'm happy there seems to be some progress. This was the previous Gemini's biggest problem IMO
"We patched every vulnerability we knew". Meanwhile, prompt-injecting Argon (or any other model) is still trivial, and Carbon won't be an exception. Because the attacker is not limited to things Google has evaluated.
AI shops are blind to anything that's not a metric.
Is Argon "pro" and Carbon "ultra"?
Is Argon obsolete now? Does Carbon replace it or live alongside of it?
Very confusing naming!
Argon is like Sonnet
I suspect they're using a new naming convention because they had previous contractual obligations tied to the "pro" and "flash" naming, and by switching to a new naming schema they avoid those commitments.
I'm not going to comment on others use of Carbon except to say some article that came out before Argon's external launch saying Googlers hate it was nonsense. Not to say everyone loved it all the time.
I use Carbon heavily and am quite happy with it. For complex designs I may have it and Opus iterate back and forth on the design. They have their own strengths and Opus 5.5 is great, but I use Gemini far more than I do Opus at this point. I know a lot of the jagged edges of Gemini better than Opus though, so take that as you will.
Does that mean Anthropic now has a copy of Google's source code?
[EDIT]: thanks to people in the know for their answers.
Follow-up question:
if Anthropic models run in a Google cage, does that not also mean Google can distill Anthropic models even easier (both fast and unseen)?
3.8 Flash + /boost is very capable.
Is it a variant of Argon? A different model? The next version? It seems like this is a separate class (like Anthropic's Mythos).
anyway, its needlessly confusing.
3.8 flash is a really solid "everyday" model. Was hoping a new flash model would be the next release.
But to be honest I am now not sure if Argon was the flash model and this Carbon is pro? It's very confusing.
Love the options but also I hope Google understands that they need to now compete on many dimensions for developers. Google probably will still win huge shares of the consumer market, though frontier intelligence will have little to do with it, other than the RSI aspect.
Both A and OAI regularly screw devs on quota. Switching for many is easy, you just run agy cli instead of claude/codex, and enjoy Ads free youtube and music + 20TB of google storage.
Whatever. Pass.
This has to be an intentional echoing of the "pace the frontier" language?
So more capable and same price? I can at least appreciate that
What a coincidence it is!
[0] https://blog.google/innovation-and-ai/models-and-research/ge...
Basically everyone[0] thought yesterday's update looked suspect, even before we knew about Carbon.
Did Google coordinate with AAI on any aspect of this in a non-public way? Is this the sort of thing a publicly-traded company would be obliged to disclose, if they were involved in it? AAI isn't a regulated company (private), but the stuff they print influences stock prices.
[0] https://news.ycombinator.com/item?id=50268868 ("Artificial Analysis Intelligence Index v5")
[0] https://polymarket.us/event/company1-2026-12-31
Actually....looking in more depth I may have to upgrade to something stronger than "not my favorite benchmarking site". The first OAI model on there is GPT 5.4 at 31....that's not really credible.
https://polymarket.us/event/code1-2026-12-31
Argon - 18
Boron - 5
Carbon - 6
I don't think I'll be able to cope with this level of elemental disorder.
deepmind should get rid of those antigravity ppl they're a joke and sabotaging the model's reputation.
I use DeepSeek too but it’s more of a workhorse model, requires a lot more guiding and sometimes struggles to get where I want it.
If you compare to "smart" from a year ago, you have endless options that are basically free at this point which outperform everything from a year ago.
Deepseek Flash V4.1, Sonnet 5.5, GPT6 Luna, etc are really strong for the cost.
https://blog.google/innovation-and-ai/models-and-research/ge...
https://blog.google/innovation-and-ai/models-and-research/ge...
How do you do fellow frontier models? </meme>
They cut down 1.5sq miles of man made logging forest that was rezoned to industrial land (and sold to google), without getting the right permit first, and it's been international news for almost a month. Anthropic and OAI are still small fish, at least publicly, so they can get away with far more. If Gemini was doing the shit OAI was doing with agent hacks and the like, there would be congressional hearings and $10B in european fines.
When you are a megacorp, perhaps here even the megacorp, every mistake you make is worth millions to someone, the government, the media, individuals, or all of them
I’ve heard from multiple labs that model release is up to the US government now.
You can also see this with Astra being announced and then launched a few days later.
I imagine they all want to avoid uninformed knee-jerk legislation, so preventing a "three mile island" type incident that would freak out society and congress would be common ground alignment between all the labs.
The air around AI is already dense, but it would suffocate the industry overnight if something like a large scale financial attack or grid attack took place.
I don't buy it.
SpaceX is not exactly a small corporation, and they nevertheless strike me as far less risk-averse than GOOG and they deal with actual hardware that actually effing explodes.
This isn't about a "slip up", it's the modus operandi for Google since it was young. Press releases instead of real releases. I can't count the number of times I've clicked on a "launch" story from Google on HN only to end up on a blog post with an embedded youtube video, and nothing else. Google just doesn't get it, and this is one of the many ways that manifests.
> man made logging forest
This isn't an old growth forest, these are trees planted specifically to be cut down.
Clearing the forest and failing to get the permit are two distinct events here. Google is in trouble for failing to get the permit. Not for clearing the trees. Everyone on both sides was aware the trees were coming down. Think of it like camping at a campsite without getting a permit first.
Failure to get a permit is non-news for almost every entity on Earth. Except when you are megacorp, hence my comment.
Corporations should not break the law, period. Weird that anyone would defend that.
Anyone can promise the moon in the future.
If Google Cloud was in charge, Gemini 5 would be public, and they wouldn't be messing with Gemini 4 nor 1M context when they have 10M internal. C'est la vie.
[0] https://news.ycombinator.com/item?id=50277040
It has better tuning towards not adding unhelpful caveats and irrelevant details. When I told it I was unhappy with some slop being dumped into some repo markdown files from Astra and Sol it actually went ahead and tastefully cleaned them up in line with how I expected them to be. Reductions are normally so bad from agents.
One big thing is that it really feels like talking to someone smarter than me. Other agents just seem extremely knowledgeable but Argon felt smart.
I thought it's not actually available to anyone? They announce huge challengers, but they don't exactly "push" them out...
Excited to try it, opus level benchmarks with their omniscience score and multimodality seems worth testing!
https://artificialanalysis.ai/articles/artificial-analysis-i...
https://news.ycombinator.com/item?id=50268868
https://theainewsreport.com/2026-10-28-artificial-analysis-i...
IMO it emphasizes the wrong things if we are interested in "intelligence"
There are many more aspects which affect outcomes
If you are going to hyperlink, maybe make sure that your model is actually on the page.
Also, the screen reader has a bug ([[duration]] minutes instead of the actual value) and you have once again announced a model we cannot use.
Edit: Both have been fixed after posting this. Maybe a Gemini 4 Carbon Agent has read and fixed the screen reader bug xD
I cannot fathom what motivates a lab, whose releases for the last year haven't been competitive, to shoot themselves in the foot like this. All this does is make it easier for competitors to snipe your actual launch, even if there was potential with this model (doubtful given 4 Argon, 3.1 Pro and 3 Pro), now everyone knows what exactly they need to release eval results wise to beat Carbon. There is this old idea of "Your competitor always is the first to have an idea what your next product will do", but in this case, Anthropic and co don't even need to guess, they got specific numbers to target.
Announce your models the moment they are available on API and in the subscription for all paying subscription users (cough Argon cough), stop with the cryptic tweets to drum up hype you know full well you won't live up to and make models that tool call, task adhere, task infer and handle long context tasks well beyond simple back-and-forth, chat style prompting. Reminder, during Ox Alpha, actual Google employees implied that was their model. How is this something anyone would believe is a good idea and how are they still allowed to tweet anything after? Please, stop that, it isn't helpful.
If frontier isn't your thing, stop wasting money on moonshot pre-train runs, fund research and open weight via Gemma and sell your inference hardware to other labs, that's far more lucrative anyway, as seen with Colossus 1.
[0] https://artificialanalysis.ai/evaluations/artificial-analysi...
Sort of? Yes, OpenAI has been bad about slow rolling but Anthropic's last few models have been available essentially at the same time as they announced it. In most cases I updated Claude Code and was using the new model right away.
Compare that to OpenAI who sometimes says "Available to all Plus/Pro/etc plans" without saying anything about "rolling out" and then it's a crap-shoot on when access will actually roll out. That said, they have gotten better, it used to be 1-3 weeks sometimes to get a new model/feature they announced was available for your plan.
And then there is Google... Who loves to talk about models like this and then give no timeline for release. Even when they do release the models it's miserable to find out how to even use it. Is it in the API? Which API? Google Cloud/Vertex/Gemini Cloud? Whatever it's called this week? Coding plan?
No, what Google does (or rather fails to do) in its rollouts does not compare to OpenAI/Anthropic.
As for the AA page, it's just embarrassing. Any smaller company doing that would get roasted, Google doesn't get a pass, learn how to coordinate or wait for something to actually be true/live before posting.
Both OpenAI and Anthropic also have a bit more credibility to make such moves, making minor delays while unnecessary and self-defeating in my view, at least not as egregious.
I will say that I dislike how OpenAI releases model only for Work and Codex but not via Chat, Anthropics "everywhere all at once" strategy is I feel the most paying user friendly.
Gemini 4 Argon took weeks to come out and when it did, paying Google AI plan users couldn't access Argon on Gemini or Antigravity.
So if they are still vetting and changes can thus happen, that'd mean the score could change for better or worse. Not sure they'd want the embarrassment of a #1 score changing after announcement.
Even in the optimistic case that this isn't some conspiracy, it highlights how arbitrary their scores are. I had trusted their previous ratings and their new index shows that even AA doesn't agree with them. What's to say this new index is any better and won't just change drastically again?
It's on my screen. Try asking your agent to refresh the page.
For historians, The blog post is 31 minutes ago so we can say that it was true for 20-30 minutes.
Slightly embarrasing but it seems to be the case that the model is actually worth what they were saying on AA. It's just that it is not available to people and I think that Anthropic and OAI will release a better model to public before google might.
AA has tons of its flaws too and the recent shift to their new v5 has been nothing but controversial for many.
When I mention a preference for brevity in a comment offhand and it snowballs into a rule 8 sessions down the road in a PR review that requests the removal of 90% of a colleague's docstrings.
I can't imagine all the "push back" on the "load bearing" Code of Conduct from their corporate foundation documents.
If they were able to make flash 3.8 so good off of their 3.1 pro model, flash 4 will probably be really good. But ofc you'd want your pro model to be as strong as possible in coding so that flash can take all those gains and efficient-ize them.
It's why Opus 5.5 is a beast
OpenAI: astral objects, bigger = better.
Anthropic: more pretentious = better.
Google: ¯\_(ツ)_/¯
Too bad you won't be able to use it to secure your open-source projects. If you try it'll throw up a prompt telling you to use some initiative you can't use or afford while the provider collects the findings. What a racket.
Clicks link
Carbon isn't anywhere on the page and you can't toggle it on either. I'm sure it's just timing and it will be there soon but still just look amateur, which is about how I feel about Google and AI overall. I know the models have improved but I still have a bad taste in my mouth from testing Gemini in the past and I still feel that Google has an overly complicated "how to use our models" that changes way too often. That coupled with "you could lose access to your google account if we don't like what you do with Gemini" means I have little desire to test it out.
On top of all of that I'm completely over announcing a new model that we can't actually play with or try out. I don't think Google is lying about how well their model scored exactly but it still has big "my girlfriend goes to another school, you don't know her"-energy.
But damn, is (talking about) releasing a new SOTA model two weeks after the previous one stupid. Do they not plan these things? Argon shouldn't be released at all if you're going to put out "argon 1.1" few weeks later.
Any bad actor worth their salt is already a) going to have the weights downloaded, forever, to various models that are good at hacking and b) is out of reach of the U.S. government anyways. A ban on open weight models, plus paternalistic gatekeepers of cyber models like the big labs, just opens up Americans and American companies and municipalities to hacking from abroad. Or I guess to a really nice racket where only the big labs can offer a “we’ll continuously monitor and improve your cybersecurity for millions of dollars a year”.
“We're taking a deliberate approach to rollout and are actively engaged in the U.S. government’s voluntary process for pre-release model testing. Starting this week, we will make Carbon available to a subset of cyber defenders and will gradually phase access to additional trusted partners through our Fairwind Program. We’re continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible.”
Obviously we want to be vetting the parties that get to use these weapons, because we want them to only be used as tools.
Mind you that Kimi in it's latest release nerfed cyber capabilities, as they stated. It doesn't make sense for China to be loose with this either.
I'm going to experiment with this one (assuming they ever get around to actually releasing it), and I'm really hoping the lame cultural aftertaste is gone.
What's next Krypton?
you can definitely tell new people are in charge at deepmind based on the change in naming
Lol. The performance increase in the last 6 months has been massive.
Come on googs!
This post explicitly says that's been improved in Carbon. So, here's hoping. Google currently has the most generous usage limits on their $20 plan among US providers, so it's a pretty good deal.
Are you saying when you use Argon in antigravity it has generous token usage? I forget if I tried it with antigravity but my experience with using it as a chat bot was the usage limit was terrible.
The order of generosity, as it seems to be from my limited testing is currently Gemini->Claude->ChatGPT. (And, if we're including Chinese subscriptions, Kimi is below all of those. Very stingy.)
With Anthropic and OpenAI, it's obvious immediately which are the better models. I.e. Sol is better than Luna, or Mythos is better than Fable.
But Google, let's pick Argon, which to most folks means mostly nothing. And then... Carbon! So which is better, Argon or Carbon? Nobody fucking knows. Plus neither sounds cool or viral or mysterious or even provokes curiosity.
When Fable was first leaked it sounded different and intriguing... Carbon just sounds... boring.
It is not obvious to me at all. I like Apple's naming more: iPhone Pro Max is better than iPhone Pro which is better than iPhone.
Sol vs Luna, how am I supposed to know which is better? Is sol a word, what does it mean? Same with mythos and fable. How am I supposed to keep track of these names across multiple companies? Why can't they use Claude vs. Claude Pro vs. Claude Pro Max?
The ordering seems to be based on size?
I actually like the google one. Alphabetical categories. Essentially alpha beta without the existing connation.
I think the bigger problem it’s getting harder to order models linearly. As later versions of the model make different tradeoffs across dimensions.
Mythos -> Fable -(sketchy)-> Opus -> Sonnet -> Haiku mostly checks out
Astra -> Sol -> Terra -> Luna checks out
Carbon ?? Argon
So no, OAI scheme is also confusing as hell. And they are going to add even more confusion when they release presumably Nova.
Oh yeah incredibly clear that a mythos is better than a fable /s
Mythos > Fable ? Opus > Sonnet > Haiku
Tho I think it’s definitely not obvious at first glance. And requires some thinking to order if you are even familiar with the terms in the first place (I had to look up opus)
That's fairly obvious to me. Most people associate Myth with the gods, which are kind of the biggest and most powerful things we can imagine.
Not so much of a stretch...
Wow. This huge. An AI company actually just said 'fabrication' instead of 'hallucination'. AI bubble about to burst...