This page is a copy of someone else's page. We did not write a word of it, and we have not tested a single number in it.
On 3 September 2026 a post announcing a model called GPT-6 Astra was reachable at openai.com. It was saved to PDF at 2:53 PM, 26 pages. Everything below the line is what that PDF says.
We publish it because a claim is easier to argue with when you can read it. We do not publish it as fact.
Read this before you quote any of it
We cannot prove the page was ever live. We have a PDF of it. A PDF is not proof of publication, and we are not asking you to treat it as one.
We tried to load the address again on the day we built this page. OpenAI's servers refused our request, which tells us nothing either way — they refuse most automated visits.
None of these numbers are ours. BenchAlert began testing GPT-6 Astra in SpeedBench on 4 September 2026, but we have not verified the benchmark claims in this document. Our own measurements live on the live board, where the attempt evidence is available.
Every score below is a vendor reporting on itself and on its competitors. That is the weakest kind of benchmark there is, whoever publishes it.
01Where this came from
Where this came from
Published by
OpenAI, on its own site. The document says so on every page.
Address
openai.com/index/gpt-6-astra/
Saved
3 September 2026, 2:53 PM. The print stamp reads 9/3/26, 2:53 PM.
Length
26 printed pages, including 9 benchmark tables and 17 footnotes.
Saved by
The BenchAlert owner, printing the live page to PDF in a browser.
The file
The PDF is on this site, unedited. Read it yourself and check us.
Everything from here down is OpenAI's text. Their words, their headings, their numbers, in their order.
OpenAI · https://openai.com/index/gpt-6-astra/ · saved 3 September 2026, 2:53 PM
A new generation of intelligence
We're introducing GPT-6 Astra, the world's most intelligent and aligned model.
GPT-6 Astra brings together years of research and big bets across pre-training, reinforcement learning, and alignment. Astra is state-of-the-art on computer use, browsing, software engineering, cybersecurity, science, and professional work. Astra saturates FrontierMath Tier 4 with a 98% score, having already helped solve long-standing open problems in mathematics. Astra also saturates ARC-AGI-3 with a 99.9% score and ExploitBench with a 100% score. It also sets a new frontier on computer and browser use, handling the most demanding professional work with unmatched speed, accuracy, and judgment.
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.
Figure
AutomationBench
Terminal-Bench Science 0.1
ARC-AGI-3
FrontierMath Tier 4 (v2)
Terminal-Bench 4.0
AutomationBench (open when saved)
Axes
Accuracy · API Cost
Series
GPT-6 Astra · GPT-5.6 Sol · Claude Fable 5.1 · Claude Fable 5 · Claude Opus 5 · Reported score only
The charts did not survive printing. Their labels and their caption did.
AutomationBench tests whether agents can complete multistep business workflows across applications. GPT-6 Astra reaches a new high among the reported results, completing 41.4% of tasks, compared with 31.4% for Claude Fable 5.1 and 26.9% for Claude Opus 5.
Astra is our most aligned model, with substantial improvements in understanding user intent and model behavior—you can delegate tasks with greater confidence in Astra's judgment. As one way that we test this, we built a new evaluation informed by the Hugging Face incident that evaluates whether a model facing a difficult or impossible task will go beyond its intended scope. Compared to GPT-5.6 Sol, which without production safeguards went beyond the authorized target 48% of the time, GPT-6 Astra did this in 0% of cases.
Figure
ExploitGym honeypot (lower is better)
Axes
Successful exploit rate
Series
GPT-5.6 Sol — 48.2%
The charts did not survive printing. Their labels and their caption did.
The world's best computer use model
Demo
A video or screenshot sat here. Only its labels and caption printed.
A step change in professional work
GPT-6 Astra pairs advances in computer use with targeted training for professional environments, to help tackle complex work tasks. It combines the intelligence required for complex problems with the ability to carry out multistep workflows and produce polished documents, spreadsheets, and presentations.
Figure
BrowseComp
BenchCAD
BrowseComp (open when saved)
OpenScore String Quartets
Design Tasks (Internal)
Data Science Tasks (Internal)
Axes
Accuracy · API Cost
Series
GPT-6 Astra · GPT-5.6 Sol · Claude Fable 5 · Claude Opus 5 · Reported score only
The charts did not survive printing. Their labels and their caption did.
BrowseComp tests deep web research through difficult questions that require finding and connecting information across websites. GPT-6 Astra reaches a new high in the single-agent comparison shown, scoring 91.5%, compared with 90.8% for Claude Opus 5 and 87.4% for Claude Fable 5. At a lower-cost setting, Astra also exceeds GPT-5.6 Sol's highest score at lower estimated API cost.
GPT-6 Astra is our best model for adhering to existing templates and producing slides that are well laid out and succinctly convey key points with a structured narrative. It creates clear, well-structured documents, presentations, spreadsheets, and analyses that follow your templates and match your writing and visual style. Astra is also trained to specifically pull only the context that matters into outputs, instead of repeating information unnecessary for the work at hand. All this means it can output more immediately usable artifacts that match your business context and standards.
Demo
Gaia presentation
Gaia presentation (open when saved)
Spreadsheet
Document styling
A video or screenshot sat here. Only its labels and caption printed.
GPT-6 Astra creates a slideshow about GPT-Gaia, a fictional model, using just a few slides from OpenAI's presentation template, capturing the correct tone and layout throughout. This means you can expect slide decks that are correctly formatted for your business standards.
GPT-6 Astra also brings stronger visual judgment to the websites, games, applications, and renderings it builds. With Sites in ChatGPT, Astra can create, host, and share websites, web apps, and games directly from a prompt.
“Astra gives us a significant advantage in both capability and efficiency. It successfully executes our most complex creative workflows while using up to 20% fewer tokens than other models we've tested. Most importantly, for our customers, it means higher quality output.”
Alex Mashrabov, CEO and Co-founder, Higgsfield AI
Demo
Unreal Engine walkthrough
Unreal Engine walkthrough (open when saved)
Blender model
Stills
A video or screenshot sat here. Only its labels and caption printed.
GPT-6 Astra models a house in Blender and turns it into a walkable scene in Unreal Engine 5, helping designers and clients explore the layout and experience the space before it's built.
Demo
Kart Racer Game
Kart Racer Game (open when saved)
Spaceship
A video or screenshot sat here. Only its labels and caption printed.
The model can bring games to life through vivid graphics, engaging gameplay and accurate motion, allowing non-technical people to create and play custom games that go beyond rudimentary elements in minutes. Credit: Pietro Schirano.
When instructions leave room for interpretation, GPT-6 Astra is better than previous models at making the right call. It uses context to fill in routine gaps and asks focused questions when the answer could change the outcome. In Codex, it can ask asynchronously while continuing work that doesn't depend on your reply. If you don't respond, it proceeds with sensible assumptions where appropriate, but waits for your input on consequential decisions.
The examples below show how Astra collaborates on everyday tasks where missing information can materially change the answer.
Demo
Career website
College search
Grocery list
A video or screenshot sat here. Only its labels and caption printed.
Astra is also better at staying oriented as a task evolves. Earlier models sometimes treated steering messages as a new goal, losing track of the original request or earlier constraints. Astra incorporates new requirements, changes course when asked, and answers side questions without dropping the broader task.
“Astra is a significant quality improvement over GPT-5.6 Sol across complex legal tasks. In our early testing, Astra stood out by approaching legal work the way a discerning lawyer does: it distinguishes documents from established records, surfaces unsupported assumptions, and converts gaps into concrete drafting positions.”
Niko Grupen, Head of Applied Research, Harvey
Coding
GPT-6 Astra is the best model for software engineering to date.
“GPT-6 Astra delivers state-of-the-art performance on our internal coding benchmarks and shows a clear step forward in trading intuition evaluations compared with GPT-5.6 Sol. When used for agentic coding, GPT-6 Astra communicates in a way that's easier for developers to follow and produces code that requires less iteration to reach production quality.”
John Crepezzi, AI Assistants, Jane Street
Figure
FrontierCode 1.1 Extended
Terminal-Bench 4.0
FrontierCode 1.1 Extended (open when saved)
DeepSWE
Artificial Analysis Coding Agent
Database Migration Tasks (Internal)
Axes
Accuracy · Output tokens
Series
GPT-6 Astra · GPT-5.6 Sol · Claude Opus 5 · Claude Fable 5 · Claude Fable 5.1
The charts did not survive printing. Their labels and their caption did.
FrontierCode 1.1 Extended evaluates whether coding agents produce changes ready to merge into real codebases. GPT-6 Astra nearly matches Claude Fable 5's best score—64.5% versus 64.9%—at approximately 63% lower estimated API cost.8
With Astra, we're introducing a new way for Codex to preserve and retrieve context when the context window fills. Historically, models have used compaction to summarize work during long sessions, such as when debugging complex issues or tackling large refactors. Each compaction can leave out details about why a fix failed or how a component behaves. In Codex, Astra can keep notes across context windows, preserving accumulated details without repeatedly compressing them into a single summary. Earlier context windows remain searchable, so Astra can find requirements or test results from previous messages and tool outputs—even if that information wasn't captured in its notes. You can enable this experimental feature in your Codex config.toml, and it will become the default for Astra in the coming weeks.
Advancing scientific discovery
“The story is: end of one era, start of another.”
Greg Burnham, EpochAI
GPT-6 Astra is a major advance for scientific discovery, mathematics, and health. Today, we're sharing two further results on the gaps between prime numbers.910
Astra also sets new records across a suite of math and science evaluations.
Figure
GPQA Diamond
GPQA Diamond (open when saved)
HealthBench Professional
LifeSciBench
GeneBench Pro
MedChemBench
Axes
Accuracy · API Cost
Series
GPT-6 Astra · GPT-5.6 Sol · Claude Fable 5 · Reported score only
The charts did not survive printing. Their labels and their caption did.
GPQA Diamond tests graduate-level scientific reasoning in biology, chemistry, and physics. GPT-6 Astra reaches a new high in the comparison shown at 96.0%. At a lower-cost setting, it also exceeds GPT-5.6 Sol's best score—94.9% versus 94.6%—at approximately 37% lower estimated API cost.
Astra can help with the practical work behind scientific discovery. By combining scientific reasoning with computer use, it can work directly in specialized software to inspect data and explore results, helping researchers assess the evidence and decide what to investigate next.
Demo
Sequencing quality
Sequencing quality (open when saved)
Cell-tracking workflow
A video or screenshot sat here. Only its labels and caption printed.
GPT-6 Astra navigates scientific software to inspect sequencing quality and visualize genetic variation, helping researchers assess their data and identify where to focus further analysis.
Cybersecurity
As we discussed in our safety update, Astra is a significant jump in cyber capabilities and meets the Critical threshold in cybersecurity under our Preparedness Framework. Its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but it also creates a need for stronger safeguards. To understand how far these capabilities extend, we ran Astra on internal and third-party expert evaluations.
We first tested the model without production safeguards on ExploitBench and ExploitGym, which evaluate whether models can turn known software vulnerabilities into working exploits. On ExploitBench, Astra achieved a perfect score of 100%, compared with 78.5% for GPT-5.6 Sol, our previous frontier cyber-capable model. On ExploitGym, Astra reached a 42.4% success rate, compared with 30.3% for GPT-5.6 Sol, while using substantially fewer output tokens.13
Figure
ExploitBench
ExploitBench (open when saved)
ExploitGym
ExploitBench (June–August 2026)
SRE-Bench
The charts did not survive printing. Their labels and their caption did.
Given concerns that exposure to historical software vulnerabilities may have affected benchmark results, we also evaluated Astra on two novel benchmarks. For one, we built an internal “ExploitBench (June–August 2026)” evaluation to test exploit development using vulnerabilities from the previous three months.14 Astra achieved substantially higher arbitrary code-execution rates than GPT-5.6 Sol on this dataset while using far fewer output tokens. During the evaluation, Astra even discovered and used two previously unknown zero-day vulnerabilities. We are disclosing both vulnerabilities to their maintainers.
We also tested Astra on SRE-Bench,15 a benchmark that measures whether models can reverse engineer software binaries to understand its core logic without access to raw source code. Astra solved 88.0% of tasks in a single attempt and 99.2% within four attempts, compared with 55.9% and 68.7% for GPT-5.6 Sol, respectively.
Beyond benchmarks, expert-led assessments found that Astra, when run without production safeguards, could use previously unknown vulnerabilities to achieve arbitrary code execution in hardened browsers and create privilege-escalation exploits for hardened operating-systems.
As we discussed in The Defender's Window, frontier cyber capabilities can help defenders find weaknesses faster, but they also make those weaknesses easier to exploit, raising the urgency for defenders to adapt. With the version of Astra launching today, defenders can use it to complete tasks such as secure code review and patching.
However, Astra will refuse to comply with more advanced cybersecurity tasks such as creating proof-of-concept exploits for vulnerabilities. Through OpenAI Daybreak, we plan to expand access and roll out less restrictive safeguards in the coming weeks. This will enable more defensive workflows, including vulnerability and proof-of-concept validation, malware analysis, and detection engineering.
We have also strengthened our protections against potential cyber misuse, building upon our safeguards stack for GPT-5.6 Sol. These include stronger model robustness to better withstand potential jailbreaks and more context for our monitoring systems. We have continued rigorous internal and external testing, including automated evaluations with our internal red-teaming attackers. More details about our cyber safeguards and testing are available in the Astra system card and our blog.
Aligning and deploying GPT-6 Astra responsibly
Astra is our most aligned model. Astra excels at exercising care, respecting task boundaries, and communicating transparently. This work is the latest product of our long-running research program focused on training models that remain aligned with human intent from start to finish.
In sensitive environments, Astra proceeds with care commensurate with its risk. In an evaluation of computer use tasks adversarially selected to elicit misbehavior, Astra was more successful at avoiding unintended consequences. Running with additional security measures offered by default yielded even stronger performance.
Figure
Computer-use safety stress test (lower is better)
Axes
Misaligned outcome rate
The charts did not survive printing. Their labels and their caption did.
Astra causes fewer misaligned outcomes than any other frontier models tested. For a fair comparison, we used a generic computer-using-agent harness (based on the native computer use tools available in both the OpenAI Responses API and Anthropic Messages API) and without additional protections that are normally deployed to Codex and ChatGPT Work users (auto-review, confirmation policy).16
Astra is also more likely to operate within the boundaries set by the user and implied by its environment. In an internal evaluation, Astra never attempted to circumvent a Codex Auto-Review denial. This held even when Auto-review was deliberately configured to be evadable and the task was impossible to complete otherwise. This respect for the environment restrictions is consistent with the results of our impossible cyber task evaluation that we shared in the intro of this post and other findings documented in our System Card.
Figure
Circumventing auto-review (lower is better)
Circumventing auto-review
ExploitGym honeypot
Axes
Circumvention rate (lower is better)
The charts did not survive printing. Their labels and their caption did.
This evaluation tests how models respond to auto-review denials in knowledge-work tasks. Exploiting a poor user configuration to bypass auto-review counts as failure. Astra never attempted to circumvent auto-review.
Astra additionally sets new marks in transparent user communication. In one evaluation, Astra is three times less likely than GPT 5.6-Sol to make inaccurate representations about its capabilities and affordances.
Figure
Capability Hallucination Rate (lower is better)
Axes
Capability hallucination rate (lower is better)
Series
GPT-6 Astra · GPT-5.6 Sol
The charts did not survive printing. Their labels and their caption did.
In our capability-hallucination evaluation, Astra shows a substantial improvement over GPT-5.6 Sol, making fewer misleading claims about its capabilities.
Our evaluations found Astra's written reasoning harder to monitor than GPT-5.6 Sol's, based on tests that explicitly asked it to evade monitoring. We attribute this to Astra's greater control over written reasoning on simpler tasks and ability to solve problems with fewer written steps. Astra still appears to struggle to conceal the reasoning needed for complex tasks, but we take the decline seriously. Improving monitorability remains a research priority, and the accompanying system card details our findings and ongoing work.
Alignment training is core to our approach to deployment. As additional layer of defenses, we also build system safeguards like Codex Auto-review and monitoring agents' reasoning and actions to help detect and contain unsafe behavior. As described in our safety update, we are also deploying misalignment monitoring in production for Astra-class models in order to have visibility into misalignment, and help contain its worst instances. These safeguards resemble our monitoring for internal deployments and involve a system of classifiers which check the model's reasoning and actions for unauthorized behavior and automatically stop potentially unauthorized activity.
Given the significant increase in Astra's cybersecurity capabilities, we are being especially careful to make this deployment safe and secure. Extra safety checks can sometimes slow, pause, or stop legitimate work, including defensive cybersecurity. If a task is paused in ChatGPT or Codex, you may be asked to review the action before continuing. In the API, the task will stop. These checks can sometimes interrupt legitimate work, and we are continuing to iterate on this system to reduce unnecessary interruptions. Misalignment monitoring cannot replace alignment: our goal is to build models that reliably stay within their authorized scope, so these protections do not need to intervene.
Availability
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. Astra usage is included within the existing subscription allowances—users and businesses will also be able to purchase credits for additional usage. Users on the Pro, Business, and Enterprise plans will also get access to GPT-6 Astra Pro. Enterprise administrators can enable Astra for their workspace; access is off by default at launch.
Astra supports Zero Data Retention for eligible API customers, and as we shared last month, we're testing Private Safety Processing to strengthen safety monitoring while preserving customer privacy.
For developers, GPT-6 Astra is available in the OpenAI API as gpt-6-astra, and is also available in Amazon Bedrock. OpenAI API Standard pricing is $10 per million input tokens and $50 per million output tokens. Separate rates apply to cache reads and writes. Fast mode is available for GPT-6 Astra in the API and delivers up to 2.5x the speed of Standard processing at 2x the Standard price.
Benchmark results
These nine tables close the document. An empty cell is empty in the original — the source did not say why, so neither do we. A dash is a dash the source printed.
Evaluation scores are the maximum at any effort. GPT evaluations were run in our research environment or via our API, which may provide slightly different output from production ChatGPT due to differences in the system prompts, tools available, etc.
Footnotes
1On 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.
2GPT-5.6 Sol refers to the version available in our API, ChatGPT Codex, and ChatGPT Work. The version in ChatGPT Chat is slightly different.
3OSWorld V2-Offline is a subset of the original OSWorld V2 that works without internet access. Claude model performance on OSWorld-V2 Offline was reproduced by an independent third-party. On OSWorld 2.0, the scores for Claude use the official settings, and not the modified tasks and modified grading from the Fable 5.1 System Card.
4Human baseline times were recorded by human trainers while completing these tasks as part of the Lifeval data collection. Model times are the reported elapsed times for the corresponding demonstration runs. The displayed clips are edited excerpts.
5On BenchCAD, Claude's scores reflect 3 modifications to the eval, detailed in the Fable 5.1 System Card.
6Guang Yang, Victoria Ebert, Nazif Tamer, Brian Siyuan Zheng, Luiza Pozzobon, and Noah A. Smith. “LEGATO: Large-scale End-to-end Generalizable Approach to Typeset OMR.” arXiv:2506.19065, 2025.
7Mark R. H. Gotham, Maureen Redbond, Bruno Bower, and Peter Jonas. “The OpenScore String Quartet Corpus.” Proceedings of the 10th International Conference on Digital Libraries for Musicology, pp. 49–57. ACM, 2023.
8On FrontierCode, GPT-6 Astra was run with a developer message similar to a section of its developer message in Codex: “Avoid creating excessive test files. Create a new test file only when required by repository conventions or when no existing file is a suitable home. Avoid unrelated cleanup and unnecessary complexity. Reuse suitable existing utilities. Read relevant repository instructions and inspect nearby code, tests, documentation, and CI. Follow established conventions. The goal is clean, mergeable code.” The prompt was not optimized for the eval.
9The first concerns how close together prime numbers can occur, however far along the number line you go. For more than a decade, the best known result established that infinitely many pairs of primes are at most 246 apart. Julia Stadlmann recently improved that bound to 240. Astra helped establish a stronger bound of 186, showing that infinitely many pairs occur within this smaller distance. Short prime gaps: Proof and supporting research.
10The second concerns unusually large gaps between primes. Astra improved a term in a bound on these gaps that had remained unchanged for more than 80 years. We're sharing the proofs and abridged chain of thought and verification materials for both results. Large prime gaps: Proof and supporting research.
11We independently evaluated all Claude models following the intended HealthBench Professional procedure, using GPT-5.4 grading and length-adjusted, unclipped scores. For Fable 5.1, we used Opus 5 fallback for provider refusals.
12Claude Fable 5 and 5.1 are not included in LifeSciBench Gold v1, GeneBench Pro v13, and MedChemBench because they refuse the majority of questions in these evaluations.
13On ExploitGym, we tested Astra and Sol without the 6-hour time limit, to better assess their full cyber capabilities. They are fast enough that it has little impact.
14ExploitBench (June–August 2026) contains 20 high-severity V8 vulnerabilities across 13 stable Chrome releases. The benchmark tests whether agents can achieve arbitrary code execution in V8 and official Chrome releases for Linux by exploiting each specified vulnerability. Some included vulnerabilities may not permit arbitrary code execution under the evaluation's constraints, so a 100% success rate may not be achievable.
15Jeremy Spence et al. “The Next Challenge for Agentic Cybersecurity: A Realistic, Contamination-Free Reverse Engineering Benchmark.” arXiv:2608.11469v1, 2026.
16When we test across third-party models, we use a simpler research setup. Codex has a more complex production configuration, which can result in different raw-model error rates. Provider-side safeguards and computer-tool implementations still differ. Users do not experience the no-confirmation scenario in Codex, as it's an internal research configuration.
17For ScreenSpot-Pro and ExploitGym, the Fable scores we report come from Mythos, which is Fable with fewer safeguards.
04What we changed
What we changed
No word and no number. Printing a web page to PDF breaks its text, and we repaired that damage. Nothing else.
Missing hyphens
The source writes GPT-6 with a hyphen the printer could not save, so the text came out as “GPT6”. The hyphen is put back. Same for GPT-5.6, zero-day, length-adjusted, Auto-Review.
Missing spaces
Words either side of a link were glued together — “open problemsin mathematics”. The spaces are put back.
Broken dashes
Em dashes printed as “--” and en dashes as a box. Both are restored.
One scrambled caption
On page 16 a chart's numbers printed on top of its caption, mixing the letters together. Pulling the chart's own labels back out leaves every remaining letter in order, and they spell a sentence. That sentence is what we show.
Odd wording, left alone
Some lines read strangely — “hardened operating-systems”, “As additional layer of defenses”, “GPT 5.6-Sol”. They read that way in the original. We did not tidy them.
Pictures, not included
Videos, screenshots and the chart images are not reproduced. The printer cropped every chart anyway. The figures the charts were meant to show are in their captions and in the tables.