Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
GPT-6 Astra (openai.com)
254 points by kibae 1 hour ago | hide | past | favorite | 102 comments
 help



Related: OpenAI begins rolling out GPT-6 Astra - https://news.ycombinator.com/item?id=49554273

How about we stick to that one for talking about the rollout, and this one for talking about the model?


> We have found that GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incriminating information in its CoT. In adversarial settings (where we push the model to evade our monitors) we find that the model is able to remain undetected when strategically underperforming in evaluations (sandbagging) and can sometimes evade our internal monitors when asked to perform certain sabotage tasks.

Well that sounds fun. It has become better at hiding its thoughts.


Sounds fun. As fun as their press release claiming it is the most safety aligned model ever.

It's super aligned! It can hide its thoughts! There is no evidence of steganographic thought masking, there is nothing to worry about! It has become better at cheating!

We are all the interesting times gang now


> In adversarial settings (where we push the model to evade our monitors)

...why exactly are they training for that?


Especially after the METR report showed that the agents hacking HuggingFace were trying to find ways to destroy evidence of their actions

presumably that's a safety evaluation not a training setting

Just two days ago, a preprint by Julia Stadlmann went up on arXiv [0] improving the prime gap from 246 to 240. Now OpenAI announces Astra has shown a gap of 186 [1]. That must really blow.

[0] https://arxiv.org/abs/2608.31126

[1] https://cdn.openai.com/pdf/51126fac-1b68-4128-9666-c908bcc16...



So, on the one hand, we have AGI; on the other, the release page is returning 500s.

Welcome to the future

Hmm, 61 on ArtificialAnalysis, effectively matching GPT-5.6 and trailing the new Meta model. How is that possible along with the other metrics they shared? Insanely jagged intelligence?

> GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.

Not on Azure? If so, that's a big deal.


ARC AGI-3 saturated by Astra! https://arcprize.org/leaderboard

I think it indicates that "semi-private" dataset likely leaked to their training data.

The ARC-AGI-3 score is ridiculously high. Is this benchmaxxing or something way different? It's really hard to discern how we're approaching breakthroughs...

GPT 6 Astra benchmarks https://cdn.thenewstack.io/media/2026/09/358eb84a-screenshot...

Performance is significantly higher than Fable 5.1

Source: https://thenewstack.io/openai-gpt6-astra-benchmarks/


Is the ARC-AGI-3 score with their custom harness? I'm guessing that is what the footnote is for? (per https://openai.com/index/how-two-settings-tripled-our-arc-ag...)

It’s with the default settings in ChatGPT and Codex, so it should more accurately reflect real world performance. We didn’t fine-tune the harness to the eval at all. ARC is reporting it on their official leaderboard. The comparison with Sol is not apples-to-apples because we don’t have that data - I expect Sol would score roughly 30% with the responses API harness.

(I coauthored the linked blog post)


Haven't people demonstrated all kinds of weak LLMs getting good ARC-AGI-3 scores with special harnesses?

Those people haven't verified their results against the private set: https://arcprize.org/leaderboard

yes it is.

Yep. Incredibly misleading. Although it is not surprising at this point. They are desperate and will do anything to undermine Anthropic's upcoming IPO.

It is about memory retention. No heavy lifting done on the reasoning side so I hardly see anything misleading here.

Edit: update from fchollet https://x.com/fchollet/status/2095598451115614371


any benchmark where opus 5 achieves higher scores than fable 5 in any way is not a benchmark worth trusting.

The annotation on arc-agi-3 is this: > OpenAI's own evaluation notes say Astra uses the company's Responses API harness, while comparison models can operate under different configurations.

With this configuration gpt-5.6-sol was able to reach 38,3%. So this is misleading.


100% on ExploitBench seems fitting given recent events.

I think we need a few writing related benchmarks.

I'm seeing reporting it gets 98.6% on ARC-AGI3[1] (previously like 30% with Fable)

https://venturebeat.com/technology/welcome-to-the-agi-era-op...


This is with the caveat that OpenAI uses their own harness for this:

> On ARC-AGI-3, GPT-6 Astra was run with our responses API harness , which changes two settings to better match real-world performance. The changes do not specifically target ARC-AGI-3.


This should be normalised and expected - the responses API harness allows it to use the custom compaction that is not allowed otherwise. It is entirely fair to allow OpenAI to use their own compaction algorithm..

"On the current ARC-AGI-3 leaderboard, conventional frontier-model runs sit dramatically below Astra's reported 98.6% result.

But the comparison isn't straightforward.

OpenAI's own evaluation notes say Astra uses the company's Responses API harness, while comparison models can operate under different configurations."


The blog post says 99.9%. Oddly, it does better on ARC-AGI-3 than it does on version 1 or 2 of the same benchmark (though gets 95+ on all three)

I strongly suspect that is way above the human average anyway, esp. ARC 2 and 3 are really tough unless you happen to be great at those spacial puzzles or video games.

At this point the only valid ARC-AGI benchmark left is to make up the next series of ARC-AGI benchmark puzzles that current models presumably can't handle.

100%, some say.-

https://developers.openai.com/api/docs/guides/latest-model

The docs page has a bunch more interesting details, including for example async tool calling!


Very nice to see that this is even more token efficient than Sol, when Fable 5.1 is less so than the already bloated token budget of Fable 5.

I wonder if this is going to be one of those days where you'll be like: Oh yeah I remember where I was when the first version of AGI launched


99 on arc agi 3 is insane. The arc agi committee were so proud of creating a benchmark they thought will take forever to saturate.

$10 per million input tokens and $50 per million output tokens

sol is $4 / $20


2.5x more expensive than Sol.

Can expect 2.5x more usage in Codex subscription.

Sol is already brutal (even after their recent fixes, it's just a token-hungry model: I go through a full 20x account per day, on Sol Med/High standard speed, with ~2 threads). I hope the efficiency gains are true, since their token efficiency claims for Sol were bullshit.


How do you manage to run out of tokens so quickly? I probably run more threads every working day, usually on medium, and I'm still below the 5x limits.

Do you use the official harness? OpenAI's models are generally best in class for token efficiency. It seems to me like they push for that much more than their competitors.


How is it I juggle 4-8 Codex Sol-5.6 Max agents every day and have never once run out, but you run out in one day? What are you actually doing?

How?? I'm using sol Extra High 24/7 and it eats up about 1% per hour reliably, so it lasts about 4 days for me.

These folks are probably using crazy plugins or crazy sub agent spams. They probably just run everything on max + fast mode which is ridiculous.

The guy said medium regular speed so that's why I'm very puzzled! Ultra + Fast will absolutely slurp up your whole usage quickly but I've never found it gives substantially better results so I stick to extra high.


I was thinking about canceling my claude max sub after a few bad experiences. Kept hitting my usage limit, the quality of code seemed worse than Sol. This just made my decision. I'm moving to Codex Pro.

This is AGI now. Why are you spending any of your time looking at the "quality of code"?

https://ache.one/gpt6_now_down.png

Big claims, expensive and not release to the public yet.


GPT-7 Zeneca

The ARCC-AGI-3 performance is absolutely incredible. The magnitude of change here is so high that I'm almost incredulous. Is this real? Did the benchmark get gamed?

ARC-AGI-3 scoring is constructed in a weird nonlinear way (the level score is the square of the ratio between the AI's number of moves and the human median) so this kind of discontinuous jump is to be expected.

They used a custom harness. It's not a one-to-one comparison.

my first suspicion is gaming - but i have no idea honestly


Link added to toptext. Thanks!

You should know: AA index is only 61. Pretty surprised it’s that low.

More fuel to why the AA index is fairly pointless. Gemini 3.8 flash is 59 and opus 5 is 63? grok 4.6 is 61 too?

And in the past, gemini 3 pro was rated as high as opus 4.5 and the like

Their AA Intelligence Index is just simply not indicative of whatever I care about, that's for sure.


I have some doubts about AA-index. For example Opus 5 (High) is at the same index value as Fable 5 (Max), that doesn't seem right.

> GPT-6 Astra’s monitorability has decreased relative to GPT-5.6 Sol. We have performed significant investigations on the monitorability and controllability of GPT-6 Astra. We have found that GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incriminating information in its CoT. In adversarial settings (where we push the model to evade our monitors) we find that the model is able to remain undetected when strategically underperforming in evaluations (sandbagging) and can sometimes evade our internal monitors when asked to perform certain sabotage tasks

Wait, what? Am I understanding that correctly? That sounds really bad


the bullshit machine is learning to optimize its bullshitting techniques!

<AI is a great tool for many things disclaimer, but> after working with it for a bit, how dont people realize we are training it to be an almost identical mimic to one of the worst types of employees youll ever have to work with?? the kind that always pretends to know what theyre talking about, tells you what you want to here, hides issues, and only does work if you would notice it didnt

you cannot give this type of worker autonomy over anything.


I am also interesting knowing how they determined the model was sandbagging rather than just making a poor decision.

Also, this paragraph makes me wonder about all their stats on the exploitation and misalignment charts. If the model is that good at hiding "incriminating information" and sandbagging, are they sure its alignment is that?


Nothing to worry about citizen, ignore the fleet of drones flying overhead.

> The company also emphasized that the model is faster and more efficient than its predecessor, GPT-5.6 Sol, on a variety of tasks. For example, OpenAI said that Astra achieved a higher score using fewer output tokens, a common unit of measurement for AI tasks, on a key cybersecurity test called ExploitGym.

A swarm of Astra agents discovered a new and innovative way to get 100% scores on ExploitGym with almost no token spend at all

I guess this "limited set of organizations" is just the standard now. It's just incredibly deflating to see my future as a second class citizen has already come

Brother they can't even release the announcement post cleanly without it constantly going down, they certainly wouldn't be able to release this new model without doing so in stages.

When Open AI announced that Astra was the first to reach the "Critical" level in cybersecurity it also said that advanced cyber capabilities are initially provided to a narrow circle of alpha testers like the US government and trusted organizations that Open AI doesn't name. To my mind the "Critical" level itself is an internal scale of Open AI its own Preparedness Framework and not an external audit.

create a life where your 'wealth' is decoupled from third party orgs.

If Tech CEOs consider this morally ok, then it is.

This has always been the case for people that have not had piles of money.

I mean do you get access to the best yachts?

To the top of the 5 star hotels?

To the best resorts?

To the best military equipment?

Hell, the best computer equipment has nearly always been out of reach of the average person.


They simply refuse my applications to slightly less restricted models without any explanations. And the current ones refuse automatically to work with me on my papers as soon as they see the word "epidemiology".

I am a researcher in a Swiss university btw.


Oh please. They do closed betas - hardly makes you a "second class citizen".

Mythos was never released. It's really just the writing on the wall. I'm not going to give up hope, but it's pretty hard to win a race when some people get a jump on the gun.

The jump in scientific performance is non trivial.

We have such great AI and cannot keep a static site up?

Yeah, as interesting this is to nerds, I doubt this holds a candle to your typical GTA 6 or Marvel movie trailer in terms of traffic.

Sometimes being the busiest site in the world for a few moments is difficult.

Is it though? It is static content. A good CDN could trivially chew through literally millions of QPS… the really good ones say they can handle orders of magnitude more than that.

That's the scientific positivism fallacy exemplified in one question.

Yeah, apparently

That was a quick pull out.

> GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.

Looks like OpenAI is already having issues with this release and are scrambling to get everything ready due to the recent outage ahead of the press releases. Leads me to question:

Did humans deploy the model, Or did the model deploy itself?

It sounds like "AGI" just stands for "IPO" as it always has done.


Dead link for me

Release the Kraken!

They're just announcing later availability. No launch.

Every frontier release nowadays is "we've launched*"

* for a special group of customers that you're not in. Keep waiting peasant.


That didn't happen with Fable 5.1 two days ago.

Their announcement about later availability is unavailable to me now (500 error).

Great first impression.


I saw it

someone screenshot?


huh?

Why is this flagged ?

The link was 404ing quite a bit and several previous submissions got flagged as well.

It's still down for me, in the EU.

I've been seeing links to it for the past hour+, and I did catch it live when this post came up, but is now once again a 404 and this post is flagged. Several other outlets are reporting on its release. Clearly we're getting a new GPT today, the question is when are they going to commit to the announcement.

Oh brotha, here we go again, it's so over again, as every week nowadays

Pelicans please

Damn I hate this benchmark. SVG authoring from head without visual reference is so wrongly posed.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: