ChatGPT Astra refers to OpenAI's newly introduced GPT-6 Astra, announced on September 3, 2026. OpenAI says Astra is its most capable model yet, with major improvements in reasoning, coding, computer use, browsing, science and professional workflows. Its API price is $10 per million input tokens and $50 per million output tokens, while Claude Fable 5.1 from Anthropic has the same headline input/output pricing but differs in capabilities and benchmark performance. 

Introduction

OpenAI has officially introduced GPT-6 Astra, the model behind the latest generation of ChatGPT capabilities, marking a major new step in the company's AI model lineup.

The announcement came on September 3, 2026, with OpenAI describing GPT-6 Astra as its most intelligent and aligned model. The company says Astra is designed for difficult, multi-step work across computer use, software engineering, browsing, cybersecurity, science and professional tasks.

The launch comes just two days after Anthropic introduced Claude Fable 5.1, creating an unusually close comparison between two newly released frontier AI models. Both models are priced at $10 per million input tokens and $50 per million output tokens through their respective APIs, but their reported strengths differ across benchmarks and use cases.

What Happened?

OpenAI announced GPT-6 Astra on September 3 and began rolling it out initially to a limited group of organizations.

The company says broader access will follow over the coming days for ChatGPT Plus, Pro, Business and Enterprise users. Astra is also being made available through the OpenAI API and cloud platforms including Microsoft Azure and AWS Bedrock.

OpenAI's developer documentation identifies the API model as gpt-6-astra.

The model has a 1.05 million-token context window and supports a maximum output of 128,000 tokens. Its listed knowledge cutoff is April 30, 2026.

Importantly, Astra is not simply positioned as a faster chatbot. OpenAI is emphasizing its ability to complete long, multi-step tasks involving computers, browsers, code, documents and professional software.

Key Details About ChatGPT Astra

Some of the most important GPT-6 Astra specifications and features include:

• Model: GPT-6 Astra
• Announcement: September 3, 2026
• API input price: $10 per million tokens
• API output price: $50 per million tokens
• Cached input: $1 per million tokens
• Context window: 1.05 million tokens
• Maximum output: 128,000 tokens
• Reasoning levels: Low, medium, high, xhigh and max
• Knowledge cutoff: April 30, 2026
• Capabilities: Coding, research, computer use, browsing, science and professional workflows
• ChatGPT availability: Plus, Pro, Business and Enterprise rollout planned
• API availability: OpenAI API, with cloud availability including Azure and AWS Bedrock.

The $10/$50 pricing is particularly important because it puts Astra's headline API rates at the same level as Claude Fable 5.1.

What Are the New Features in GPT-6 Astra?

OpenAI's biggest focus is computer use.

Astra can perform tasks such as filling online forms, updating CRM records, organizing calendars, conducting online research and working with document editors. OpenAI also says the model can analyze scientific data, create plots, build websites and perform frontend quality-assurance checks.

Another major improvement is its ability to handle changing instructions.

OpenAI says Astra is better at maintaining the original objective when users introduce new requirements or ask side questions. In Codex, it can also preserve information across context windows instead of relying entirely on repeated summarization.

For professional work, OpenAI highlights Astra's ability to create documents, spreadsheets and presentations while following existing templates, formatting requirements and visual styles.

This could make the model particularly useful for people who need finished work products rather than simply text responses.

Coding and Software Engineering

Coding is another central focus.

OpenAI describes GPT-6 Astra as its strongest software-engineering model to date. Its published benchmark results show Astra scoring 57.9% on Terminal-Bench 4.0, compared with 55.8% for Claude Fable 5.1 in the comparison presented by OpenAI.

However, benchmark results vary by test.

On DeepSWE v1.1, for example, Astra scored 74.1%, while Fable 5.1 scored 67.4%. On FrontierCode 1.1 Extended, Astra scored 64.5%, compared with 63.6% for Fable 5.1.

These figures come from OpenAI's published evaluations, so they should not be treated as an independent universal ranking.

GPT-6 Astra vs Claude Fable 5.1

Anthropic launched Claude Fable 5.1 on September 1, 2026, just two days before OpenAI announced Astra. Anthropic describes Fable 5.1 as its most capable generally available model for coding and knowledge work.

The two models have remarkably similar headline API pricing:

Anthropic additionally lists cache reads at $0.25 per million tokens, saying this is 75% lower than Fable 5's cache-read price. The company estimates the change can reduce costs for typical workloads by about 25% and highly agentic workloads by up to approximately 45%.

The benchmark picture is mixed.

OpenAI's comparison gives Astra higher results than Fable 5.1 on Terminal-Bench 4.0, AutomationBench and FrontierMath Tier 4. But Fable 5.1 scores higher than Astra on Humanity's Last Exam with tools and on Artificial Analysis's Intelligence Index shown in OpenAI's table.

That means there is no single benchmark that proves one model is universally better.

What Officials and Organizations Said

OpenAI describes Astra as its most capable and aligned model and says it represents a major improvement in computer use, professional work, coding and scientific applications.

The company also highlighted external feedback. Jane Street said Astra showed state-of-the-art performance on its internal coding evaluations, while Cognition said it was integrating Astra into its Devin development environment.

Anthropic similarly positions Claude Fable 5.1 around long-running coding and knowledge-work tasks. The company says the model can operate across multiple applications, recover from failures and work asynchronously with limited supervision.

Safety Is a Major Part of the Astra Launch

The technical improvements come with significant safety considerations.

OpenAI says GPT-6 Astra reached its Critical threshold for cybersecurity capability under its Preparedness Framework. The company reports that Astra achieved 100% on ExploitBench when tested without production safeguards, compared with 78.5% for GPT-5.6 Sol.

OpenAI says it has therefore strengthened monitoring, model protections and deployment safeguards.

Astra will refuse some advanced cybersecurity requests in its initial deployment, while OpenAI says it plans to expand defensive cybersecurity capabilities through its Daybreak program.

Anthropic has also emphasized safeguards for Fable 5.1, particularly around cybersecurity, biology and chemistry. Its Fable 5.1 and Mythos 5.1 models share the same underlying model, but different safeguards and access policies apply.

Why This Matters

The significance of ChatGPT Astra goes beyond another model-number upgrade.

The industry's focus is increasingly shifting from models that simply answer questions to models capable of taking actions.

Astra's ability to interact with websites, applications, software and professional tools points toward AI systems that can complete workflows rather than merely generate individual responses.

The price comparison with Claude Fable 5.1 is also important for developers. Both companies are charging $10 per million input tokens and $50 per million output tokens, so the practical choice may increasingly depend on task performance, token efficiency, tool integration, reliability and workflow requirements rather than headline price alone.

What Happens Next?

GPT-6 Astra is currently rolling out gradually.

OpenAI says broader access is planned for Plus, Pro, Business and Enterprise customers, while API access is also being expanded.

For developers, the next important question will be how Astra performs in real-world applications outside controlled benchmark environments.

The competition with Claude Fable 5.1 is likely to remain particularly significant because both models are aimed at long-running professional, coding and agentic workloads.

For users, the practical winner may depend less on which model has the highest single benchmark score and more on which system performs the specific jobs they need reliably, quickly and economically.

FAQs

What is ChatGPT Astra?

ChatGPT Astra refers to OpenAI's new GPT-6 Astra model, introduced on September 3, 2026. It is designed for advanced reasoning, coding, computer use, browsing, science and professional workflows.

How much does GPT-6 Astra cost?

The OpenAI API price is $10 per million input tokens and $50 per million output tokens. Cached input is priced at $1 per million tokens.

Is GPT-6 Astra available in ChatGPT?

OpenAI says GPT-6 Astra is initially rolling out to a limited group and will become available to ChatGPT Plus, Pro, Business and Enterprise users over the coming days.

What is the GPT-6 Astra context window?

GPT-6 Astra has a 1.05 million-token context window and supports up to 128,000 output tokens, according to OpenAI's developer documentation.

Is Claude Fable 5.1 cheaper than GPT-6 Astra?

Not on standard input and output API pricing. Both are listed at $10 per million input tokens and $50 per million output tokens. Anthropic does, however, offer Fable 5.1 cache reads at $0.25 per million tokens.

Which is better, GPT-6 Astra or Claude Fable 5.1?

There is no reliable single answer. OpenAI's published benchmark table shows Astra ahead on several coding, computer-use and scientific tests, while Fable 5.1 leads on some other evaluations. The best choice depends on the workload.

What are the biggest GPT-6 Astra features?

The major improvements include computer control, complex multi-step workflows, software engineering, browsing, professional document creation, scientific work and improved handling of changing instructions.

When was GPT-6 Astra released?

OpenAI announced GPT-6 Astra on September 3, 2026. Initial access began with a limited set of organizations, followed by a planned broader rollout.

Sources of the Content

This article was researched using primary company sources wherever possible, particularly OpenAI and Anthropic documentation.

• OpenAI — GPT-6 Astra announcement: official launch details, features, benchmarks, availability and safety information.
• OpenAI Developer Documentation: official GPT-6 Astra API specifications, token pricing, context window and model capabilities.
• OpenAI Safety Overview: official information about Astra's cybersecurity capabilities and safety measures.
• Anthropic — Claude Fable 5.1: official pricing, capabilities, availability, benchmarks and safeguards.
• Anthropic Newsroom: official September 1, 2026 announcement of Claude Fable 5.1 and Mythos 5.1.

Important accuracy note: “ChatGPT Astra” is commonly used as shorthand for the GPT-6 Astra launch, but GPT-6 Astra is the official model name used by OpenAI. The benchmark comparisons above are based primarily on the companies' published evaluations and should not be interpreted as an independent universal ranking.