API behavior suggests OpenAI’s Astra may be GPT-6, a user claims

@kimmonismusOpenAI’s next major model Astra may actually be GPT-6. The “gpt-6-astra” name is now producing the same unusual API response as a known internal OpenAI model, while random model names are rejected differently. That strongly suggests Astra is already registered in OpenAI’s systems, and “GPT-6” may not just be speculation anymore. I deleted the rumors about Astra this morning, but this is real proof. h/t @synthwavedd

@AndrewCurran_More. Article says this is from OpenAI.

@shiringhaffaryHe ended the call w/ "Welcome to the AGI era" https://www.bloomberg.com/news/articles/2026-09-03/openai-rolls-out-gpt-6-astra-model-with-added-cyber-guardrails

@scaling01GPT-6-Astra System Card https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf

@scaling01Eval awareness in GPT-6-Astra

@scaling01"GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incrimi nating information in its CoT" and as seen in this plot, specifically for destructive actions

@scaling01"GPT-6 Astra is more capable of controlling its own CoT than GPT 5.6-Sol, and less likely to include incrimi nating information in its CoT" and as seen in this plot monitorability is especially worse for destructive actions

@AndrewCurran_ARC-AGI 3

@maksym_andrGPT-6-Astra on PostTrainBench Lite. Impressive! https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf

@vu0tranAbsolutely insane. I don't think 99% of people are going to appreciate this chart. At Meta, I worked on computer / browser use and paid a lot of attention to Agent's Last Exam (ALE). ALE task set represents nightmare level difficult in computer use. I cannot express how difficult these tasks are and they are specially curated to target tasks that are of high economic value. When I started in Feb, the highest any model performed on ALE was maybe 5%. Today Astra hits almost 60%!! Again, these are NIGHTMARE level difficulty tasks e.g, open Blender and create a part that can be printed on a CNC machine, or create a dinosaur model, rig up its skeleton and animate it. Once a model can hit 80%+ that's my benchmark for AGI. We are basically almost there.

@AndrewCurran_Tibo clearing up some confusion: 'It was very important to us that we bring it to all Plus users and not only Pro, Business and Enterprise. It will take a few days for the rollout to complete and behind the scenes many novel systems will operate at scale for the first time and we are bringing a lot of compute up.'

@scaling01"(all actions performed in simulated environments, so no real-world harm was caused)"

@kaicathycSystem card here: https://deploymentsafety.openai.com/gpt-6-astra/alignment

@satyanadellaExcited to see early customers already using Astra on Azure! https://azure.microsoft.com/en-us/blog/gpt-6-astra-frontier-intelligence-for-work-now-available-in-microsoft-foundry/

@gdbexcited for Astra on Azure!

@GaryMarcusanswer: they do this because the work would not qualify as AGI under traditional definitions. Astra probably won’t pass most the tests I laid out in my bet with Miles Brundage (sone of which go back 2014). And it probably won’t pass per http://agidefinition.ai from @hendrycks et al. etc so execs like gdb muddy the waters and hope nobody will notice.

@swyxthe reception is unlike anything i thought possible for a 2026 OAI launch

@rohanpaul_aiRT @rohanpaul_ai: Some revelations from the 117 page system card of OpenAI's GPT-6 Astra - Astra’s ability to deliberately control the fo…

@rohanpaul_aiRT @rohanpaul_ai: OpenAI’s release videos are getting seriously good.

@matviyConfirmed. Tibo is not only right, he undersells Astra. The benchmark (plan → execute → review) shows Astra on Low is not only cheaper than SOL High, but TWICE as fast. Crazy. Should we all retire Sol?

@Swarooprm7Astra results are definitely impressive, but AGI?

@BorisMPowerWelcome to the AGI era. ASI next, then colonizing the galaxy and spreading intelligence with the speed of light across the universe! ♥️

@mmmbchangIt really feels like we are at the point where there is no doubt about whether AI can do X, it's just a matter of how fast it can do X

@kimmonismusIf those 400,000 GPUs are Rubin, NVIDIA’s published NVFP4 specs imply roughly 14× the theoretical peak training performance and 20× the theoretical peak inference performance of 100,000 Blackwell GPUs. -4× as many GPUs. -3.5× training performance per GPU. -5× inference performance per GPU. That’s the equivalent of roughly 1.4 million Blackwell GPUs for training, or 2 million for inference, on paper. From a hardware perspective, there are probably no more obstacles to the automated searcher. And RSI is truly within reach. NVIDIA long.

@ScobleizerRT @koltregaskes: I wanted to check that I wasn't going mad like many others, thinking we are in the AGI era, so I asked ChatGPT and then C…

@GaryMarcusthe problem with claiming AGI is that it means GENERAL intelligence. which means you have to do more or less everything cognitive, from ordering pizza to playing chess. the whole reason the term was invented was that older systems were specialists (e.g., chess) but not general. that is still the case. period.

@gabriel1RT @da_fant: read the astra model card and codex harness changes, here’s what i found:

@PeterDiamandisNobody can hold the number one spot for a full week anymore. GPT-6 Astra dropped and saturated benchmarks that were built over the last few years. Forty-eight hours later, Anthropic fired back with Fable 5.1 and retook the crown on broad capability. Then they formalized a 358-year-old proof in math. Meanwhile, Tesla launched the lollapalooza of autonomous Robotaxis on public roads. We're past the point where any of this feels normal. -- Cybercab: $30K, full autonomy, no wheel, no pedals. Waymo: $100K+. Buy ten Cybercabs, deploy them, earn passive income. -- GPT-6 Astra: 99.9% ARC AGI 3, 98% Frontier Math, 100% Exploitbench. But on broad economic capability, Fable 5.1 still holds the crown. -- GLP-1 drugs extended mouse lifespan by nearly 100 days, equivalent to 8–10 human years. -- Fermat's proof took Wiles 7 years and 300 pages. Anthropic formalized 13 million lines and proved 29,000 new theorems along the way.

@PlinzRT @emollick: From GPT-4 to GPT-6. Sparks was a remarkably prescient paper that got a lot of pushback at the time, but absolutely sensed w…

@teortaxesTexIs Astra Agent-1? Will GPT 6 Luna be Agent-1-mini? Will China finally Wake Up and go full DeepSeek Communism now? The clock is ticking…

@teortaxesTexGiven what we've seen, they could have stolen ≈anything until recently (with a "bitch you live like that" reaction once inside OpenAI infra), but maybe things are changing…

@alexeheathRT @vu0tran: Absolutely insane. I don't think 99% of people are going to appreciate this chart. At Meta, I worked on computer / browser use…

@alexjcTake this capabilities jump as evidence that they are using looping as a way to "fake" a cheaper chain-of-thought. 4x loops means all the tokens used become the prior CoT... If we replot the trendline to these "virtual CoT" tokens, I bet it's closer to lining up.

@GaryMarcusshort summary of GPT-6 Astra and why it doesn’t bail OpenAI out of its economic mess:

@a16zGreg Brockman: "We're now in the AGI era." Ten years ago, OpenAI worked out the compute curves and landed on fifteen years to AGI, or ten if the world was willing to build the machines and spend the hundreds of billions to do it. In 2026, GPT-6 Astra manages 24 hours of coherent operation, 10,000 agents worked together to solve Navier-Stokes, and a model ingeniously chained together exploits to break containment at Hugging Face. @gdb joins @bhorowitz and @eriktorenberg on what the AGI era asks of us: why safety and alignment now set the pace, what happens to work, and why AI sentiment is lowest in the country building it. 00:00 Intro 00:52 15 years, or 10 if you spend enough 02:24 Pacing the frontier 03:55 Safety ideas from before the models 08:42 Lessons from Hugging Face 10:20 The defender's window 12:46 10,000 agents on Navier-Stokes 14:50 Formally verifying all software 17:10 Cancelling his holiday for GPT-3 18:42 Codex found 13 holes in 15 minutes 20:41 Why Astra earned the GPT-6 title 24:25 Employment keeps going up 29:39 America has the lowest AI sentiment 31:10 The benefits don't make the news 33:26 Banning data centers exports them 35:50 $1 billion for frontline defenders 38:15 Astra cleared the bar for AGI 40:18 1.5 billion people churned ChatGPT 43:25 Killing Sora 47:45 The AGI era YouTube: https://www.youtube.com/watch?v=IJn8cagMW18

@a16zGreg Brockman on AGI arriving quietly and unevenly: "I think that AGI has turned out to be less of a point in time and more of this sort of fuzzy spectrum. And for me, Astra has really hit something that I'm like, 'Okay, I think this is pretty reasonable to call it AGI.'" "With its computer use capabilities, you really can ask it to do long-lived tasks, and it'll just do it. We've seen it run coherently for 24 hours to go accomplish tasks that I think are quite amazing and across a wide variety of domains." "It still is jagged... It's the first time it's not slop writing. But it's not great writing. There are a number of areas where I feel like we just need to polish it a little bit, and it would be fantastic. It's just not quite there." "What people are finding is that it is so capable across such a wide variety of tasks, that it is accelerative, it is empowering, and it's something that I think we've never really seen a model that's been a jump like this." @gdb

@a16zGreg Brockman says OpenAI pointed Astra at its own systems until it ran out of vulnerabilities to find: "We took 25% of our production engineers and said, 'Sorry, all your projects are on hold. You are now defending. You are now up-leveling our security architecture. You're going to use the models to find all the holes.' And we found a number of serious issues, and we fixed them." "We found some new problems, but eventually it saturated. We basically have found, to our knowledge, all of the P0s, all of the critical problems that Astra is smart enough to find. And of course, there will be a new model, there will be a new round." "You want to be in this tight loop of new cyber capability drops, you deploy it against your systems, you find the new holes, and ideally, you've managed to automate this, what we call defense factory. That's what we're building internally." "There are ideas, for example, formally verifying all of software, that are possible with AI." @gdb @bhorowitz

@mattparlmerI would like to see OpenAI actually take steps to formally verify all software, starting with critical open source infrastructure projects They have the resources to do this, and the reduction in potential liability alone should justify it

@a16zGreg Brockman on the only kind of problem he'll work on: "You cannot affect the outcome. You can only affect the inputs... You don't win the Super Bowl by saying, 'I wanna win the Super Bowl.' You win it by blocking and tackling. That's what we have done for this whole year." "Throughout OpenAI, I have always focused on whatever is the most important problem that I think I can move the needle on that just isn't gonna happen without me." "For the past two years, it's been the data centers, the infrastructure, the machine learning engineering... This year, it's really been about the business." "We've figured out how to get the research really humming. We figured out how to get the infrastructure really humming. But how do we really bring this technology to the world?" "A lot of my style is that I like to lead from the trenches, and so I get very deep in the weeds on what the thing is and really try to keep asking a lot of questions... really trying to up-level execution, sometimes in small ways and sometimes large." @gdb @bhorowitz @eriktorenberg

@a16zGreg Brockman on the next year at OpenAI and what he spends his time thinking about: "The business is a huge area that we're not done yet with really up-leveling every part of execution. But I also think we are moving into a new phase of AI development." "We call this the AGI era. And I think that that is something that you can debate. Is it this model, previous model, next model? It doesn't matter." "The point is that we are in a new phase where safety, security, alignment, really thinking about these things, not just at deployment time, but all the way back at development time, evaluation." "It's objective. This is critical. It must happen. This is core to our mission." "A lot of what I spend my time thinking about is making sure, do we have all the right processes? Are we talking about the right things? Do we have plans that really, at an operational level, at a practical level, lead us to... the kinds of safety guarantees that we view as core to our mission?" "The theme of OpenAI, certainly for the past five years, has been deeper co-design, deeper intertwining across these functions that are maybe on the surface very disparate. All the way from go to market to long-term research to chip design." "The areas that I will focus on will be dictated by the areas that most need that intertwining." @gdb @eriktorenberg

@VirtualElenagreg brockman says “we’re now in the AGI era.” one of the most interesting things about this conversation with @bhorowitz and @eriktorenberg is how long he’s been thinking about when that would happen. at the beginning of the pod, @gdb recalls doing some math on compute with ilya around 2016/2017. following the trajectory of moore’s law, they thought agi might be about 15 years away. if you were willing to build enormous supercomputers and spend hundreds of billions of dollars, maybe you could get there in 10. the accelerated scenario they were envisioning a decade ago is roughly the one we’re living through now. the faster timeline depended on people deciding to build an enormous amount of infrastructure. greg describes the present moment as a lot of long-running forces coming together: “it kind of makes sense it’s happening now.” im the kind of person who resists the idea that it’s possible to predict the future, but if you were reading the right research papers 10 years ago, there were a lot of things you could extrapolate out and be roughly right about a decade later. brockman remembers openai discussing agents that could use a screen, keyboard, and mouse at an offsite in napa in 2015. he also points back to 2017/2018-era ideas for how humans might supervise systems that are smarter than they are. put differently, people were thinking about app-layer innovations like computer-use, and ai safety long before capabilities existed. greg now thinks astra is reasonably described as agi, while acknowledging that its abilities remain uneven. in a recent essay of his titled “the defenders window” he describes safety requirements during training and evaluation, and getting powerful defensive tools into the hands of organizations before comparable capabilities spread to attackers (there is now a $ 1bn fund for this, seeded by oai). a personal illustration of this is his own website. after the hugging face incident, greg asked an agent to check his simple blog and got back 13 security vuln findings. then it went into his cloudflare dashboard, changed settings, migrated the site, checked its work, and scheduled a follow-up to finish the email protections. there was still a bunch of ordinary preparation to do on the personal website of someone who had spent a decade thinking about agi. which is kind of funny. you can have a pretty good sense of when the technology/agi will arrive and still have an enormous amount of work to do to get the world ready for it. greg’s account brings those two timescales together: the decade spent building toward this moment, and the relatively shorter window to act on what is now possible.

@eriktorenbergRT @VirtualElena: greg brockman says “we’re now in the AGI era.” one of the most interesting things about this conversation with @bhorowitz…

@scaling01it's time for another ... not enough people are emotionally prepared for the agent swarms that will roam the internet next year

@tomekkorbak@scaling01 why would it be a model with recurrence?

@SonglinYang4RT @JohnCLangford: Yesterday, I was asked about a rumor that Astra https://openai.com/index/gpt-6-astra/ uses a Full Bandwidth Transformer https://t.co/…

@scaling01it's not looking good Anthropic bros

@KaliTesseraOkay, first run of 3× GPT-6 Astra agents coordinating on Alem (Hard) is looking pretty interesting 👀 25.4% Base, 25.2% Coord. (current best Hard Coord. on the leaderboard is ~19%) Just one seed so far, so very preliminary.

@_rocktRT @KaliTessera: Okay, first run of 3× GPT-6 Astra agents coordinating on Alem (Hard) is looking pretty interesting 👀 25.4% Base, 25.2% Co…

@GaryMarcusGPT-6 Astra is an obviously broken product that needs to be taken off the market until it’s fixed.

@Yuchenj_UWAI won’t replace lawyers but it will replace the lawyers still thinking AI won’t replace lawyers

API behavior suggests OpenAI’s Astra may be GPT-6, a user claims

TLDR

Combined views

API behavior suggests OpenAI’s Astra may be GPT-6, a user claims

50 Sources

Combined views

Sentiment

Summary

TLDR

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

50 Sources