alt Hacker NewsGemini 4 Carbon
Comments
What CLI harness should be used with this? Is antigravity usable these days via CLI?
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.
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"?
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.
Yes but has it broken out of its sandbox and hacked anyone yet?
well at the very least it will encourage anthropic to launch their new opus
Interesting - at time of writing, Carbon is claimed to be #1 on AA, but AA has not updated yet.
Excited to try it, opus level benchmarks with their omniscience score and multimodality seems worth testing!
> 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.
[0] https://artificialanalysis.ai/evaluations/artificial-analysi...
What makes me most excited is that we probably get fable 5.5 soon with the IPO. Which is going to surpass this again. But cool to see progress!
The fact that frontier models actively attempt to subvert evaluation harnesses (0.4% attempt rate still remaining) remains fascinating and slightly unsettling.
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.
So, next is Gemini Xenon or Gemini Helium?
At this pace it will soon be Gemini 5 cornholio. We’ll soon need an llm to explain between llm’s…
When will Boron be released??? That's an important question here.
Is it possible these days to create gacc not tied to phone number or identity?
A rose is just a rose…
Does it rank on FelonyBench, though?
I wonder what kind of problems are they internally testing these frontier models if not math problems?
Google is putting much higher priority on frontier models over Flash models, even though they have TPUs.
Leave it to Google to require a Chemistry degree to be able to differentiate their LLMs.
OpenAI: astral objects, bigger = better.
Anthropic: more pretentious = better.
Google: ¯\_(ツ)_/¯
I'll skip this and tune in to Gemini 94 Plutonium
Where are Gemini 4 Nitrogen and Gemini 4 Azote?
I always forget they have a model by the time I actually get access.
> 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.
i'm only here for the pelican
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.
I am on a paid plan for Google's AI offering, and I did just about ten prompts to Argon before I hit my 5 hour limit. I'm not holding my breath this will be a model that is practical to use now either.
seriously "big if true"... looking forward to seeing non benchmark tasks and seeing if this actually outperforms opus 5.5
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 are overall ahead of Astra.
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.
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.”
Argon -> Carbon
What's next Krypton?
you can definitely tell new people are in charge at deepmind based on the change in naming
Just cancelled my Anthropic subscription. Looking forward to this.
They managed to piss off all classes of their customers with their latest Argon release. They removed old flash models and gemini pro without actually giving Argon to everybody. Then they released the model to only paying customers to the exclusion of free plans or promotional deals like that comes with their Pixel phones. They staggered the release of Argon and didn't give it to Pro and Ultra users at the same time. Pro users when they got it received an extra usage pool for it (didn't help - they can't do anything with their token quota) while Ultra users have a single pool shared with Gemini Flash. I am not even sure if Antigravity got it yet. Literally nobody is happy about the release and they don't seem to care at all.
I tried previous Google models before and they were ok but I prefer to stick to Claude or Astra these days.
Maybe I am just paranoid but at least if OAI or Anthropic ban my account I am not at the risk of losing access to my email address too.
You never know with Google and that is a problem.
i dont know how to contribute to this platform.
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.
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.
Well, they won the race on confusing the shit out of us with model names.
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/)
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.
I wonder if this is finally a frontier model or just dogshit(i.e. behind open models) like all the past 3 models they released so far.
why come google is not coming up with navier stokes level proofs?
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.
Argon has been so disappointing that I don't know what to think. It feels like an older-generation model, except for its world knowledge.
> Coming soon
Come on googs!
Is Carbon to Argon what Fable is to Opus?