Artificial Intelligence
GPT-6 Astra: What OpenAI's New Model Means in India
The first model OpenAI has classified Critical under its own Preparedness Framework. That classification, not the benchmarks, is the story.
By Maya Iyer · Updated · 10 min read
Short answer
Is GPT-6 Astra available in India?
OpenAI released GPT-6 Astra on 3 September 2026, first to business users in its Daybreak programme, with a stated rollout to ChatGPT Plus, Pro, Business and Enterprise plans, the OpenAI API and Amazon Web Services in the days following. It is the first model OpenAI has designated Critical under its Preparedness Framework, on the company's own finding that it can autonomously find previously unknown security weaknesses and build working exploits against defended systems without a human directing each step. The publicly available version is built to refuse advanced offensive security tasks, including generating proof-of-concept exploits. For most Indian users the practical effect is a stronger general model on the tier they already pay for; for anyone doing legitimate security work it means refusals on tasks earlier models would attempt.
Our practical verdict
OpenAI released GPT-6 Astra on 3 September 2026, and the notable thing about it is not a benchmark score. It is that OpenAI classified its own model Critical under its Preparedness Framework, the first time it has placed a model in that band.
The stated basis for that classification is specific: the company found the model can autonomously discover previously unknown security weaknesses and build functional exploits against well-defended systems, without a human directing each step. Whatever you think of a lab grading its own release, the classification is a public commitment, and it shaped how the model shipped.
It shipped first to business users in Daybreak, OpenAI's programme for approved testers of its most capable models, with a stated rollout to ChatGPT Plus, Pro, Business and Enterprise, the API and Amazon Web Services in the days following. That is a paid-tier release. No India-specific pricing was published alongside it, so the reasonable assumption is that existing tier prices apply until OpenAI says otherwise, and the reasonable action is to check your own account rather than trust a launch date in an article.
Shortlist
Recommended options to compare
Use this as a starting list, then compare live India prices and warranty before buying.
Pick 1
What Critical actually means
It is a rating on OpenAI's own Preparedness Framework, not an outside audit. OpenAI's stated basis is that the model can autonomously discover unknown vulnerabilities and build functional exploits against well-defended systems without step-by-step human direction. Astra is the first model the company has placed in that band.
Pick 2
Who got it first
Business users inside OpenAI's Daybreak programme, which gives approved testers access to the most capable models. Consumer and API access was stated to follow within days of the 3 September announcement rather than simultaneously.
Pick 3
Which tier you need
OpenAI named ChatGPT Plus, Pro, Business and Enterprise, plus the API and Amazon Web Services. That means paid tiers. No India-specific pricing was announced alongside the launch, so assume the existing tier prices until OpenAI publishes otherwise.
Pick 4
What it will refuse
The public build declines advanced offensive tasks, proof-of-concept exploit generation named specifically. This is the trade for shipping a Critical-rated model publicly at all, and it is worth knowing before you plan work around it.
Pick 5
If you do security work in India
OpenAI said it will loosen restrictions for vetted defenders through Daybreak, covering vulnerability and proof-of-concept validation, malware analysis and detection engineering. That is a vetting process, not a setting, so budget time for it rather than expecting access on signup.
Pick 6
The benchmark to treat carefully
Astra was reported to score 100% on ExploitBench. A perfect score on a security benchmark tells you the model is capable on that benchmark. It does not tell you it is right on your codebase, and no benchmark substitutes for review of what it produces.
Which option should you choose?
What Critical actually means
It is a rating on OpenAI's own Preparedness Framework, not an outside audit. OpenAI's stated basis is that the model can autonomously discover unknown vulnerabilities and build functional exploits against well-defended systems without step-by-step human direction. Astra is the first model the company has placed in that band.
Who got it first
Business users inside OpenAI's Daybreak programme, which gives approved testers access to the most capable models. Consumer and API access was stated to follow within days of the 3 September announcement rather than simultaneously.
Which tier you need
OpenAI named ChatGPT Plus, Pro, Business and Enterprise, plus the API and Amazon Web Services. That means paid tiers. No India-specific pricing was announced alongside the launch, so assume the existing tier prices until OpenAI publishes otherwise.
What it will refuse
The public build declines advanced offensive tasks, proof-of-concept exploit generation named specifically. This is the trade for shipping a Critical-rated model publicly at all, and it is worth knowing before you plan work around it.
If you do security work in India
OpenAI said it will loosen restrictions for vetted defenders through Daybreak, covering vulnerability and proof-of-concept validation, malware analysis and detection engineering. That is a vetting process, not a setting, so budget time for it rather than expecting access on signup.
The benchmark to treat carefully
Astra was reported to score 100% on ExploitBench. A perfect score on a security benchmark tells you the model is capable on that benchmark. It does not tell you it is right on your codebase, and no benchmark substitutes for review of what it produces.
Choice IQ pick
Need the faster shortlist?
Start with our recommended options, then compare the tradeoffs that matter for your budget and workflow.
See top picksHow to decide
Pick the option around the job you need done. This is the fastest way to avoid paying for something that looks impressive but does not change your real workflow.
| Situation | Best starting point | Final check |
|---|---|---|
| What Critical actually means | It is a rating on OpenAI's own Preparedness Framework, not an outside audit. OpenAI's stated basis is that the model can autonomously discover unknown vulnerabilities and build functional exploits against well-defended systems without step-by-step human direction. Astra is the first model the company has placed in that band. | Use this as a shortlist, then verify the final details before committing. |
| Who got it first | Business users inside OpenAI's Daybreak programme, which gives approved testers access to the most capable models. Consumer and API access was stated to follow within days of the 3 September announcement rather than simultaneously. | Use this as a shortlist, then verify the final details before committing. |
| Which tier you need | OpenAI named ChatGPT Plus, Pro, Business and Enterprise, plus the API and Amazon Web Services. That means paid tiers. No India-specific pricing was announced alongside the launch, so assume the existing tier prices until OpenAI publishes otherwise. | Use this as a shortlist, then verify the final details before committing. |
| What it will refuse | The public build declines advanced offensive tasks, proof-of-concept exploit generation named specifically. This is the trade for shipping a Critical-rated model publicly at all, and it is worth knowing before you plan work around it. | Use this as a shortlist, then verify the final details before committing. |
Read the editorial notes
The publicly available build refuses advanced offensive security tasks, with proof-of-concept exploit generation named. OpenAI has said it intends to relax those limits for vetted defenders through Daybreak for Frontline Defenders, a stated one billion dollar commitment aimed at helping defenders use frontier capability to protect essential services, covering vulnerability and proof-of-concept validation, malware analysis and detection engineering. If you run security work from India, note that this is a vetting process rather than a toggle, and plan for the lead time.
For the large majority of Indian users - people drafting, researching, coding and summarising - none of that is the relevant part. The relevant part is that the model behind the tier you already pay for is changing, and the behaviour you built habits around may not be identical. The refusals are the most likely place to notice it, because the boundary of what counts as security-adjacent is drawn by the model rather than by you.
One number is worth handling carefully. Astra was reported to score 100% on ExploitBench. That is a result on a benchmark, and a perfect score on a security benchmark is not a statement about your systems. It is also worth holding alongside the fact that the same capability is why the model is rated Critical in the first place - the score and the safeguards are describing the same thing from opposite directions.
The practical step, as with every model change: take the three tasks you actually rely on, run them before assuming they behave the same, and pay attention to refusals rather than only to output quality. That is a twenty-minute check, and it is the difference between finding out now and finding out in front of a client.
Decision shortcut
Still comparing options?
Use the table above to shortlist your best fit, then check related picks, tools, and buying guides before you make the final call.
FAQ
Is GPT-6 Astra available in India?
OpenAI stated a rollout to ChatGPT Plus, Pro, Business and Enterprise plans, the API and Amazon Web Services in the days following the 3 September 2026 announcement, without naming India separately. OpenAI's consumer tiers are generally available in India, so the practical expectation is access on the paid tier you already hold, arriving on OpenAI's rollout schedule rather than on a separate India date. Check your own account rather than relying on a launch article, including this one.
What does the Critical classification mean for a normal user?
For everyday drafting, research and coding help, very little changes beyond a more capable model. The Critical rating describes cyber capability specifically, and OpenAI's response to it is the safeguards in the public build. The effect you are most likely to notice is refusal on security-adjacent requests that earlier models would have attempted, which is a deliberate design decision rather than a fault.
Should an Indian business switch to it?
Switching model is not the decision worth agonising over; the model on your existing paid tier will change under you regardless. The decision worth making is whether any workflow you run depends on behaviour that has just changed, particularly anything security-adjacent that may now be refused. Test your three most important real tasks against the new model before assuming continuity, which is the same advice this site gives on every model change.
Features, model access, limits, privacy controls, and India billing can change. Choice IQ evaluates practical workflow fit and recommends checking the provider's current official terms before paying.
AI Tools Editor
Maya reviews AI products, productivity systems, and automation workflows with a focus on practical adoption.
Keep reading
Related artificial intelligence picks
Artificial Intelligence
What 'Unlimited' Means on an AI Plan
No AI plan is unlimited. The word means a cap you are unlikely to hit under normal use, a fair-use clause that defines normal, and a smaller model waiting when you exceed it. Here is where the actual limit lives.
Comparison
AI API Pricing 2026: What Every Model Actually Costs
Every current model's token pricing in one table, including the promotional rates that expire and the long-context rates that catch people out.
Comparison
Claude Opus 5 vs GPT-5.6: Which Flagship Should You Pay For?
The two flagship models compared on what they cost, what they are measurably better at, and why most people should buy neither.
Artificial Intelligence
Model Deprecation: What Happens When the Model You Built On Is Retired
Every model you build on will be retired, usually with months of notice on the API and none in the consumer app. The prompt that worked will not quite work on the replacement. Here is how to make that a scheduled migration instead of a surprise.
Tool
AI Prompt Library Builder
Create a reusable prompt template for a team workflow, including variables, checks, and ownership.
The best choice is rarely the product with the longest feature list. It is the one you will still trust and use six months from now.
How Choice IQ evaluated this guide
Choice IQ evaluated GPT-6 Astra by what OpenAI's Critical designation under its Preparedness Framework actually asserts, which access tiers and regions were named at launch, what the public build refuses and why, the Daybreak vetting route for security work, the limits of a benchmark score as evidence, and what an Indian user on an existing paid tier should test before assuming continuity.
Weekly intelligence
Get smarter AI tool picks weekly.
One useful email with practical comparisons, refresh alerts, and decision frameworks for people who do not want another noisy newsletter.