ChatGPT Astra: How to Access OpenAI GPT-6 Astra, Pricing, API and AGI

Astra is OpenAI’s flagship model for complex, end-to-end work, with major improvements in computer use, coding, research and professional work. It can use tools, work across files and software, and continue through more steps before asking for further instructions. [1]

OpenAI launched the model on September 3, 2026. For people searching for “GPT 6”, “ChatGPT6”, “Astra 6” or similar terms, the important distinction is that the official model name is GPT-6 Astra, while ChatGPT is the product through which users access different models and experiences.

ChatGPT Astra: Price, Access and Capabilities at a Glance

The practical answer is that access depends on your plan, while developer use is billed separately. Plus includes Astra in Work and Codex, eligible Pro plans can use GPT-6 Pro in Chat, and Business and Enterprise access depends on plan and workspace permissions. [4]

Access by plan

Key specifications and standard price

For short emails, simple summaries or high-volume routine jobs, a lighter model can still be the better choice. Astra usage makes most sense when the task is valuable enough to benefit from deeper reasoning, tools, large amounts of context or autonomous execution.

What Is GPT-6 Astra and How Does It Relate to ChatGPT?

GPT-6 Astra is OpenAI’s most capable model for difficult end-to-end work; ChatGPT is the application that can provide access to it. The Astra model combines reasoning with tools for computer use, research, document creation, coding and other complex tasks. [1]

A GPT is a generative pretrained model. Users do not train a new artificial intelligence system themselves. They provide a prompt, files, information and instructions to a model that has already been trained through techniques including pre-training and reinforcement learning.

The practical difference is continuity. Instead of dividing a complex job into many isolated requests, the model can retain more context, work through a workflow and complete several actions before returning for further direction.

ChatGPT itself is the interface and product. Depending on the plan, the underlying experience may use different models and tools for conversational assistance, longer agentic work or software development.

Is Astra Available Now and How Can You Access It?

Yes, but access is not identical across every account. GPT-6 Pro, powered by Astra, is currently available for eligible Pro, Business and Enterprise plans, while Plus includes access through Work and Codex. [4]

How to get access

1. Sign in to the correct account.
2. Check the model selector in Chat.
3. Open ChatGPT Work if it is available to your plan.
4. Use Codex for software-development tasks.
5. Update the ChatGPT app to the latest desktop version.
6. If you use Codex CLI, use version 0.153.0 or newer.
7. Enterprise users should check workspace model-access permissions.
8. For developer access to GPT-6, check whether the model is enabled for your project.

Using Codex and ChatGPT with your account uses the allowances associated with that plan. Whether you use ChatGPT or Codex, heavier reasoning, large files and long-running jobs can consume the allowance faster. [4]

How Can You Use Astra in ChatGPT?

The model is most useful when you give it an objective rather than treating it as a simple question-and-answer chatbot. A knowledge worker can use it to research a market, analyse several documents, prepare a presentation and check the finished result as parts of one job.

Practical uses include research, coding, data analysis, spreadsheet work, document creation, marketing analysis, planning and software testing. The model can also work with tools rather than only returning text.

For example, a team could ask it to investigate competitors, organise the findings, identify gaps, create a structured report and prepare a presentation. That is closer to an end-to-end workflow than a sequence of unrelated prompts.

This distinction matters because not every job benefits equally from a more powerful system. The value is highest when fewer handoffs, fewer retries or more autonomous execution materially reduce the cost of completing the task.

What Are the Key Features and Capabilities of GPT-6 Astra?

The main capability shift is from answering questions toward operating tools and completing larger pieces of work. OpenAI describes Astra as state-of-the-art across computer use, browsing, software engineering, cybersecurity, science and professional work. [1]

Computer use. It can observe an interface, browse websites, click controls, complete a form, install software and troubleshoot problems visible on screen when the product and permissions support those actions.

Research and information work. It can combine sources, analyse files and preserve context through longer jobs instead of requiring the user to repeatedly restate the goal.

Documents and spreadsheets. It can work with a spreadsheet, presentation or source document while following templates and formatting requirements.

Coding. It can inspect a codebase, write and test software, diagnose failures and continue through longer development tasks.

Tool use. Developer features include web search, file search, code interpreter, computer use, hosted shell, function calling and other supported tools. [2]

OpenAI describes this as a new frontier in computer and browser use. The useful point for businesses is not the marketing phrase itself, but the move from assistance toward execution.

How Much Does Astra Cost in ChatGPT and Through the API?

Inside ChatGPT, cost depends on the subscription and its allowance. Developer pricing is usage-based. The OpenAI API standard price is USD 10 per million input tokens, USD 1 for cached input and USD 50 for output. [2]

A token is a unit used to measure model input and output. The relationship between tokens and words varies by language and content, so token counts should be treated as a billing and context measure rather than a direct word count. [3]

The model has a 1.05 million context window and can generate up to 128,000 output units. Prompts above 272K input are charged at higher rates. Batch and Flex cost 50% of Standard, while Fast mode costs twice the applicable rate. [2]

OpenAI has also announced availability through Microsoft Azure and AWS Bedrock. [1]

For a company, price per completed task is usually more useful than price per unit. A higher-cost model can still be economical if it reduces retries, supervision and manual work.

AI Performance: ARC-AGI-3, OSWorld 2.0 and ExploitBench

Published evaluation results show large gains in several areas, but no single benchmark proves that one model is best for every job. Each test measures a specific capability and depends on the evaluation harness and settings used.

The computer-use result is especially practical. In OpenAI’s latency simulation, the newer model scored 72.6% at roughly 40 minutes per task, while GPT-5.6 Sol scored 65.7% at roughly 75 minutes. [1]

The key takeaway: the numbers show where capability has improved, but the most useful evaluation for a business is still a representative task measured for quality, time, cost and required human intervention.

How Does Astra Compare With GPT-5.6 Sol, Claude Fable 5.1, Claude Opus 5 and Gemini?

There is no universal winner. Different systems lead on different tasks, and the right choice depends on the work, cost and level of supervision required.

OpenAI’s own comparison includes Fable 5.1, Claude Fable 5, Claude Opus 5 and Gemini 3.8 Flash. For example, its Artificial Analysis Intelligence Index figure is 61.2 for Astra compared with 60.9 for Sol and 65.7 for Fable 5.1, while other evaluations favour Astra much more strongly. [1]

An independent benchmark should therefore be used as evidence about a particular capability, not as a universal league table. Artificial Analysis also evaluates model performance and cost independently of the vendors. [6]

Can GPT-6 Astra Perform 3D Modeling, Coding and Scientific Discovery?

Yes. OpenAI demonstrates the model across 3D work, software development, engineering and scientific research, although human review is still necessary. [1]

For 3D computer graphics and computer-aided design, BenchCAD tests whether a model can reconstruct three-dimensional objects from multi-view rendering by generating CAD code. Astra reached a 95.9% geometric-overlap score in OpenAI’s comparison.

The company also shows the system building a house in Blender and turning it into a walkable Unreal Engine 5 environment. That kind of workflow can support architecture, product visualisation, graphics, engineering and design prototyping.

For software engineering, Terminal-Bench 4.0 evaluates complex terminal tasks including system configuration and data analysis. A capable coding agent can inspect a codebase, execute software and verify its own changes rather than only suggesting code.

In science, Terminal-Bench Science tests research workflows involving code, simulation, model fitting and data analysis. The launch materials also include examples in mathematics and scientific software. Agents’ Last Exam and Humanity’s Last Exam provide additional measures of complex tool-assisted work.

These examples do not mean the first output is automatically production-ready. They show that the system can take on a broader range of technical work before specialist review is required.

What Are the Risks and Safety Considerations Associated With GPT-6 Astra?

The main risk is the same capability that makes the model useful: it can take more actions with less human intervention. OpenAI classifies it at the Critical level for cybersecurity capability under its Preparedness Framework. [5]

Critical does not mean normal use is inherently unsafe. It means that with sufficient tools and access, the model has advanced computer security capabilities, including the ability to identify previously unknown vulnerabilities and develop ways of exploiting them.

There are also broader agent risks. A system that can work for longer may perform more incorrect actions before a person notices. Monitoring also becomes harder when capability increases faster than human supervision.

Practical safeguards

• grant minimum permissions;
• log actions and tool use;
• require approval for consequential operations;
• separate critical systems;
• restrict access to sensitive data;
• review outputs before external publication or execution;
• define clear stop conditions.

A safeguard should sit outside the model wherever possible. Businesses should not rely solely on the system explaining its own reasoning to determine whether an action is safe.

Is GPT-6 Astra Artificial General Intelligence (AGI)?

There is no independent consensus that the model constitutes artificial general intelligence. Its results have intensified the debate, but exceptional performance on ARC-AGI-3, mathematics or computer-use tests does not by itself establish human-level general intelligence.

OpenAI has described the release in unusually ambitious terms, while independent commentators have been more cautious. Gary Marcus, for example, regarded the launch as a potentially genuine advance while questioning whether benchmark saturation demonstrates general intelligence. [7]

For organisations, the label matters less than practical performance: what work it can complete, how reliably it can do it, how much supervision it requires and what happens when it makes a mistake.

Are Astra and Google’s Project Astra the Same Thing?

No. They are separate technologies from different companies. Google’s Project Astra is a DeepMind project exploring multimodal assistants that can perceive and understand their surroundings.

OpenAI’s release focuses on reasoning, tool use, coding, research and autonomous work. The shared name can also produce unrelated search results for Astra Space and SES Astra, neither of which is connected to the AI model covered here.

Are Users Leaving ChatGPT Because of Astra?

The available reporting does not support a simple claim that users are leaving because of the new model. There has been real frustration around rollout, access and usage, but demand has also been extremely high.

Some paying users criticised the staggered launch and the amount of allowance consumed by demanding tasks. Separate reporting documented complaints about access and changes in perceived behaviour after release. [9]

At the same time, demand became strong enough that OpenAI temporarily paused new sign-ups for its USD 200 Pro tier because of infrastructure pressure. That points to a capacity and rollout problem rather than evidence of a broad user exodus. [8]

The useful distinction is between dissatisfaction with a rollout and rejection of the product itself. They are not the same thing.

When Should You Use Astra?

Use it when the task is complex or valuable enough that better reasoning, computer use or longer autonomous work can justify the additional cost.

Good planning starts with the objective, budget, risk and required quality. More autonomy does not automatically mean better automation.

Frequently Asked Questions About Access, Plans and Pricing

Access is the most common source of confusion: having a paid account does not mean every model appears in every part of the product. The plan, workspace settings and whether you are using Chat, Work or a developer environment all matter.

Is Astra available for free?
There is no documented general Free access to the flagship model. Developer Free-tier access is also listed as unsupported. [2]

Is it included with Plus?
ChatGPT Plus includes the model in Work and Codex, subject to current allowances and availability. [4]

What is Astra Pro?
The official Chat experience is GPT-6 Pro, powered by Astra, for eligible higher-tier plans.

Can I use it through the OpenAI API?
Yes. Developers can select the model in supported API requests, subject to account access and rate limits. [2]

What is the context window?
The developer specification lists a 1.05M context window and a maximum output of 128K.

What is the knowledge cutoff?
The published cutoff is April 30, 2026. More recent information requires tools, browsing or context supplied at runtime.

Can it use a computer?
Yes. Computer use is one of the supported tools and one of the areas highlighted most heavily in OpenAI’s launch materials.

Is Fast mode available?
Yes in supported API contexts, at twice the applicable rate. OpenAI’s current developer guidance notes that Fast mode is unavailable with EU data residency. [2]

Learn to Apply Advanced AI to Real Work

The useful skill is not always choosing the most powerful system. It is knowing which work to delegate, what context to provide and where human judgment still matters.

The Founderz AI & Innovation Program, developed in collaboration with Microsoft, covers practical artificial intelligence, automation, agents, workflows and applied tools for professional environments.

References and Sources

The specifications, current access rules, pricing and safety claims in this guide are based primarily on first-party OpenAI documentation. That includes the launch announcement, model specification, current account-access guidance and safety material.

Independent sources are used for comparison, interpretation and reporting on the rollout rather than as substitutes for official product specifications. This separation matters because availability and pricing can change rapidly after launch.

1. OpenAI – official launch announcement
   Capabilities, evaluations, availability, Azure, AWS Bedrock, Blender, Unreal Engine 5, science and coding.
   ↑ back to top
2. OpenAI Developers – model specification
   Pricing, context, output limits, supported tools and Fast mode.
   ↑ back to top
3. OpenAI Help Center – token guidance
   ↑ back to top
4. OpenAI Help Center – Work, Codex and current access
   Plan availability, allowances, desktop requirements and Codex CLI requirements.
   ↑ back to top
5. OpenAI – safety overview
   Critical cybersecurity classification and safeguards.
   ↑ back to top
6. Artificial Analysis – independent model evaluation
   Independent performance and cost comparison.
   ↑ back to top
7. Gary Marcus – independent analysis
   Commentary on capability, benchmarks and the AGI debate.
   ↑ back to top
8. TechCrunch – demand and Pro subscription pause
   Reporting on infrastructure pressure following launch.
   ↑ back to top
9. The Verge – rollout and user frustration
   Reporting on access complaints and the staggered rollout.
   ↑ back to top
10. Google DeepMind – Project Astra
    Official information on Google’s separate Project Astra research.
    ↑ back to top