Quick Answer
GPT-6 Astra is OpenAI’s new frontier model for agentic work, but access and economics vary by product and plan. In ChatGPT, availability is still rolling out, while API access starts at $10 per million input tokens and $50 per million output tokens. The real test is whether Astra reduces retries, token use, and completion time enough to justify that premium.
OpenAI launched GPT-6 Astra on September 3, settling one question immediately: the new frontier model is Astra, not the rumored GPT-5.7. However, the name is the least interesting part of this launch. If you are evaluating Astra for production work, you probably care about three things:
- Can you actually use it yet?
- How much will it cost once you can?
- What should you give it first to see whether the upgrade is worth paying for?
Here’s what we know as of September 9, including the rollout catches, ChatGPT plan pricing, API economics, rate limits, and five jobs that should reveal what Astra can actually do.
Do You Have the Access Yet?
It depends on where you are trying to use Astra.
OpenAI is still rolling Astra out across different products and account types, so availability in ChatGPT Chat, ChatGPT Work, Codex, and the API does not necessarily arrive at the same time.
In regular ChatGPT, GPT-6 Pro, powered by GPT-6 Astra, is rolling out to the $100 and $200 Pro tiers, Business, and Enterprise. Plus, users do not currently get GPT-6 Pro in ordinary Chat, but GPT-6 Astra is rolling out to Plus through ChatGPT Work and Codex. Enterprise availability also depends on workspace model-access permissions.
For API developers, OpenAI now documents gpt-6-astra directly. There is currently no separately documented Astra Pro API model ID or separate Astra Pro API price.
Therefore, if Astra is missing from one model picker but available in another OpenAI product, that does not necessarily mean there is a problem with your account. OpenAI explicitly says rollout timing can differ between Chat, Work, and Codex. Astra also requires Codex CLI 0.153.0 or newer.
OpenAI has also designated Astra as its first model to cross its Critical cybersecurity capability threshold under the Preparedness Framework. That matters for the broader significance of the release, but staged availability is the more immediate issue for most users.
How Much Does GPT-6 Astra Cost in ChatGPT?
GPT-6 Astra pricing depends on how you access the model. ChatGPT users pay through their subscription plan, while developers using the OpenAI API are billed separately based on token usage.
The current plan picture looks something like this:
| ChatGPT Plan | Current U.S. List Price | GPT-6 Astra Relevance |
|---|---|---|
| Free | $0 | No standard GPT-6 Astra access |
| Go | $8/month | No standard GPT-6 Astra access |
| Plus | $20/month | Astra rolling out in ChatGPT Work and Codex; no GPT-6 Pro in regular Chat |
| Pro | $100/month | GPT-6 Pro powered by Astra; about 5× the Plus usage allowance |
| Pro | $200/month | Same core Pro capabilities with about 20× the Plus usage allowance |
| Business Standard | $25/user/month or $20/user/month billed annually | GPT-6 access with a more limited Astra allowance than Business Premium |
| Business Premium | $125/user/month or $100/user/month billed annually | Higher Astra usage allowance |
| Enterprise | Contract pricing | GPT-6/Astra access depends on workspace permissions and contract settings |
OpenAI also publishes separate Batch queue limits, ranging from 1.5 million tokens at Tier 1 to 15 billion tokens at Tier 5. Because rate limits can change, production teams should check the current Astra model page and their account limits page before sizing a deployment.
The same model documentation confirms Astra’s 1,050,000-token context window, 128,000-token maximum output, and April 30, 2026 knowledge cutoff. It also confirms Standard API pricing of $10 per million input tokens, $1 per million cached-input tokens, $12.50 per million cache-write tokens, and $50 per million output tokens.
What about ChatGPT pricing in India?
OpenAI supports localized ChatGPT billing in INR. It also currently supports UPI for the Go and Plus plans in India, although not for Pro or Business. App Store and Google Play subscriptions can have their own locally displayed checkout prices. API credit purchases remain USD-only.
Therefore, Indian users should check the final ChatGPT checkout price rather than simply converting the U.S. monthly subscription price at the current exchange rate.
How Much Does GPT-6 Astra Costs?
For API users, Astra standard price looks something like this:
| Models | Input/ 1M | Cached Input/ 1M | Output/ 1M |
|---|---|---|---|
| GPT-6 Astra | $10 | $1 | $50 |
| GPT-5.6 Sol | $4 | $0.40 | $20 |
| Astra premium | 2.5× | 2.5× | 2.5× |
Astra also charges $12.50 per million cache-write tokens. Batch and Flex processing cost 50% of Standard rates, while Fast mode costs 2× the applicable rate. Regional and third-party cloud pricing can differ.
For Indian teams, the OpenAI API list price does not automatically increase because the account is billed from India. OpenAI applies a 10% uplift when an eligible regional-processing endpoint is used, but India currently supports regional data storage rather than in-country API processing. OpenAI models on Amazon bedrock are billed through AWS, so the rates would differ from OpenAI’s direct pricing.
Astra has a 1.05-million-token context window and a 128,000-token maximum output. However, requests above 272,000 input tokens are charged at 2× the input and cache rates and 1.5× the output rate for the entire request. Still, price per token is not the right way to judge an agentic model. You are buying completed work, not tokens.
OpenAI reports Astra at 57.9% on Terminal-Bench 4.0 versus 37.3% for Sol, with about 9% lower estimated cost per task in its tested configurations. Artificial Analysis found Astra used about one-third as many tokens as Sol on its Coding Agent Index and cost about the same per task while scoring two points higher. However, on its broader Intelligence Index, Astra was 75% more expensive per task. Hence, your workload matters.
What are GPT-6 Astra’s API Rate Limits?
Price is only one production constraint. Throughput matters too.
OpenAI assigns GPT-6 Astra API limits according to your API usage tier. The Free API tier does not support Astra.
| API Usage Tier | Requests/Minute | Tokens/Minute | Batch Queue Limit |
|---|---|---|---|
| Free | Not supported | Not supported | — |
| Tier 1 | 500 | 500,000 | 1.5M tokens |
| Tier 2 | 5,000 | 1M | 3M tokens |
| Tier 3 | 5,000 | 2M | 100M tokens |
| Tier 4 | 10,000 | 4M | 200M tokens |
| Tier 5 | 15,000 | 40M | 15B tokens |
OpenAI says usage tiers increase as organizations send more requests and spend more through the API.
The important number depends on your application. A workflow sending a few very large agent requests may hit the tokens-per-minute limit before the requests-per-minute ceiling, while a high-volume application making small calls can hit RPM first.
Therefore, if you are moving a production workload from Sol to Astra, do not evaluate only cost per successful task. Check whether your current Astra tier can sustain the concurrency and token throughput the application requires.
Third-party providers such as Azure or AWS can have separate quotas, deployment rules, and regional limits.
GPT-6 Astra: 5 Things to Try First
Astra is more interesting when you ask for tools, adapt, and finish a job. Don’t start with write me an email; cheaper models already do that well.
The costs below are illustrative, not measured Astra traces. They use OpenAI’s $10/$50 API rates and a September 4 USD/INR mid-market rate of about ₹94.47 per dollar. Actual bills will vary with reasoning, retries, caching, and tool use.
1. Give it a messy CSV
Use a file with broken dates, duplicates, inconsistent labels, and suspicious totals. Ask Astra to clean it, validate the result, and create charts. At 15,000 input tokens and 6,000 output tokens, the illustrative cost is about $0.45, or ₹43.
OpenAI reports 40.9% for Astra versus 30.5% for Sol on its internal Data Science Tasks evaluation. That is a vendor result, so use it as a reason to test your own spreadsheet, not proof of superiority.
2. Point it at a failing repo
Give Astra the repository and a success condition. Let it inspect files, patch code, run tests, and iterate. At 80,000 input and 25,000 output tokens, the illustrative cost is about $2.05, or ₹194.
This is the first serious test anyojne should run because independent testing suggests coding is where lower token use can offset much of Astra’s higher unit price.
3. Give it three conflicting documents
Ask for one reconciled table showing agreed facts, conflicts, unresolved issues, and a source for each entry. At 60,000 input and 8,000 output tokens, the illustrative cost is $1, or about ₹94.
First thing to watch is provenance. Can Astra preserve where each claim came from instead of smoothing disagreements into a confident summary? You will have the answer right in front of you.
4. Give it a dashboard workflow end to end
Set the outcome rather than the clicks. Ask it to update records, export a report, compare totals, and flag anything that does not reconcile. At 40,000 input and 18,000 output tokens, the illustrative cost is about $1.30, or ₹123.
OpenAI reports 72.6% for Astra versus 65.7% for Sol on OSWorld 2.0 and says Astra completed those simulated tasks in about 47% less time. Those are vendor figures; nevertheless, computer use is one of the clearest capabilities to test yourself.
5. Give it a runbook to execute
Take a staging runbook, define permissions and stopping conditions, and require Astra to record changes and verify completion. At 60,000 input and 25,000 output tokens, the illustrative cost is about $1.85, or ₹175.
This exposes the assistant-versus-agent distinction. Knowing the steps is one thing; executing them and verifying the outcome is another.
Where the Costs Bite?
Agentic tasks are loops. The model observes, reasons, acts, checks the result, and repeats. Consequently, Astra’s 2.5× token premium can compound across a trajectory.
| Same Agent Trajectory | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|
| 200K input | $2.00 | $0.80 |
| 50K output | $2.50 | $1.00 |
| Total | $4.50 / ~₹425 | $1.80 / ~₹170 |
That is roughly ₹255 more for the same token volume. However, identical trajectories are exactly what Astra is supposed to reduce through fewer retries and shorter reasoning paths. The underlying rates come directly from OpenAI’s current model pricing.
Caching can narrow the gap too. On a later run, if a reusable 160,000-token prefix is already cached, the simplified Astra example falls to about $3.06, or ₹289. The initial cache write is billed separately at $12.50 per million tokens.
Therefore, I would track cost per successfully completed task, not cost per prompt.
For a fair pilot, run the same 20 to 50 representative tasks through Astra and your current model. Track success rate, retries, total tokens, tool calls, completion time, and cost per successful run. That will tell you far more than a leaderboard score.
What We Still Don’t Know About GPT-6 Astra
OpenAI has not published a firm general-availability date beyond saying Astra will expand over the coming days. Product, regional, and cloud availability are still evolving.
ARC Prize independently reports Astra at 99.9% on ARC-AGI-3 Semi-Private with OpenAI’s Provider Adapter harness, but 62.7% with ARC Prize’s provider-neutral Standard harness. ARC Prize also says Astra used fewer actions than the median tested human on 96% of levels.
This is enough to not treat one launch-day score as the final verdict. The question is simple: Does Astra finish your real work more reliably, and does it save enough retries to justify the 2.5× token price?
If the answer is no, compare cloud GPU pricing before deciding whether an open model is the better long-term option.
What to Run When the Economics Don’t Work
Not every workload needs Astra.
- Stay on GPT-5.6 Sol if your workflow already succeeds reliably. It has the same 1.05-million-token context window and 128,000-token maximum output and currently costs $4/$20.
- Wait for your tier if Astra access has not reached you yet rather than jumping immediately to metered API usage.
- Consider open weights when privacy, control, or sustained utilization justifies the infrastructure.
The third option is open weights. Qwen3.8-27B is the cleaner single-GPU candidate. It has 27B parameters, an Apache 2.0 license, and native context up to 262,144 tokens. Its standard Hugging Face repository is 55.6 GB, while the official FP8 repository is 30.9 GB.
Consequently, FP8 can fit within the 48GB memory of an NVIDIA L40S, although runtime and KV-cache memory still matter.
DeepSeek V4 is a different deployment class. V4 Flash has 284B total parameters, 13B active parameters, a 1M context window, and an official repository around 160 GB. Therefore, it is not a normal one-card deployment.
Its hosted V4 Flash API, meanwhile, starts at $0.22 per million cache-miss input tokens and $0.66 per million output tokens off peak. For sporadic traffic, hosted inference may make more sense.
We would recommend that you self-host for privacy, control, or sustained utilization and not because weights are downloadable. If you are considering that route, start with our best open-source LLMs guide and then check GPU page against the token spend you just calculated.
Frequently Asked Questions
GPT-6 Astra costs $10/M input tokens, $1/M cached input, $12.50/M cache writes, and $50/M output tokens through the OpenAI API. ChatGPT subscription usage is billed separately.
Developers can use Astra through the OpenAI API with the model ID gpt-6-astra. In ChatGPT, availability depends on your plan and whether you are using Chat, Work, or Codex.
Not officially. Astra performs strongly on advanced benchmarks, including ARC-AGI-3, but benchmark performance alone does not establish AGI. OpenAI defines AGI more broadly as highly autonomous systems that outperform humans at most economically valuable work.
ChatGPT Astra generally refers to GPT-6 Astra being used within ChatGPT. Eligible users may encounter Astra through GPT-6 Pro, ChatGPT Work, or Codex depending on their plan.
Yes, but mainly through ChatGPT Work and Codex as rollout reaches the account. GPT-6 Pro powered by Astra is not currently included in regular Chat for Plus users.
No. ChatGPT subscriptions and OpenAI API billing are separate. Using gpt-6-astra through the API is charged independently of your ChatGPT plan.
GPT-6 Astra supports a 1.05-million-token context window and up to 128,000 output tokens. Requests above 272,000 input tokens are subject to higher rates.
Astra is more capable on demanding agentic workloads, but it also costs more per token. The better comparison is cost per successful task, including retries, tool use, completion time, and human review.
