GPT-6 Astra: Release, Features, Pricing and Availability

What is GPT-6 Astra?

GPT-6 Astra is the flagship model in OpenAI's current lineup. OpenAI positions it for the hardest end-to-end work, including complex reasoning, software engineering, web research, document creation and computer use. Its API model ID is gpt-6-astra.

The emphasis is broader than producing a strong single answer. Astra is designed to carry a task through multiple steps, use tools, respond to changing instructions and keep track of a large amount of working material. That makes the release especially relevant to developers building agents and to professionals working with long documents, large codebases or research-heavy projects.

GPT-6 Astra headline specifications

• Context window: 1,050,000 tokens.
• Maximum output: 128,000 tokens.
• Knowledge cutoff: April 30, 2026.
• Input: text and images.
• Output: text; direct audio and video output are not supported.
• Reasoning levels: low, medium, high, xhigh and max.
• API support: Responses API, Chat Completions and Batch.
• Fine-tuning: not supported at launch.

A 1.05-million-token context window allows an application to provide an unusually large set of documents or code in one working context. It does not guarantee perfect recall of every detail, so important results still need verification and focused prompts.

The most important new features

Async tool calling

Astra can continue useful work while an application runs a slow function or custom tool. Developers mark an eligible tool with async: true, execute it in their own application and return the result later with the original call ID. This can reduce idle time in workflows that depend on several independent services.

Mid-turn steering

Users can send a correction or new requirement while Astra is already working. With the Responses API over a WebSocket connection, the model can preserve completed work and incorporate the new instruction into a continuation. This is useful when a long task changes direction before it finishes.

Reasoning effort can change during a conversation

A configuration_update input item can raise the reasoning level for a difficult step or lower it for a routine follow-up. The conversation can preserve its cached prompt prefix, helping applications adjust effort without rebuilding the full request.

A complete agent toolset

Through the Responses API, Astra supports web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. It also retains structured outputs, streaming, programmatic tool calling, multi-agent orchestration, prompt caching, persisted reasoning and compaction.

What GPT-6 Astra can do in practice

For a general user, the clearest improvement should appear in demanding tasks that combine several kinds of work. A single project might require Astra to read a long brief, research current information, analyze files, draft a report and revise it after feedback. The model's large context and steering features are built for that kind of sustained workflow.

• Research: browse sources, compare evidence, organize findings and draft a cited report.
• Coding: inspect a large repository, edit files, run tools and verify a change across multiple steps.
• Documents: analyze long reports or collections of files and produce structured outputs.
• Computer tasks: work across browsers and supported professional software when an application provides the necessary tools and permissions.
• Visual understanding: analyze image input alongside text, although the model does not accept audio or video directly.

These capabilities depend on the product and tools surrounding the model. Selecting Astra does not automatically give every application access to a browser, computer or private files.

Availability and API pricing

OpenAI describes the launch as a staged rollout. Enterprises in the Trusted Access Program are first, with access through the API and Plus, Pro, Business and Enterprise plans due in the following days. The official rollout notice does not promise that every eligible account will receive access at the same moment.

Standard API token prices

• Input: $10 per 1 million tokens.
• Cached input: $1 per 1 million tokens.
• Cache writes: $12.50 per 1 million tokens.
• Output: $50 per 1 million tokens.

Prompts longer than 272,000 input tokens are billed at twice the input and cache rates and 1.5 times the output rate for the full request. Batch and Flex processing are listed at 50% of Standard rates, while Fast mode is twice the applicable rate. Tool calls can add separate charges.

For API developers, Astra's higher per-token price makes testing important. OpenAI says the model can use fewer output tokens on some evaluated tasks, but the real cost per completed task will depend on the prompt, reasoning level, tool usage and amount of context supplied.

Limitations and trade-offs

• The model does not support a none reasoning setting; low is the lightest option.
• Fine-tuning is not supported.
• Audio and video are not supported as direct model modalities.
• Fast mode is unavailable with EU data residency.
• Its April 2026 knowledge cutoff means current claims still require search or another up-to-date source.
• A large context window can raise latency and cost, particularly beyond the 272K-token pricing threshold.

Astra can still make mistakes. High-stakes facts, generated code, calculations and source citations should be checked. Give tools only the permissions required for the task, especially when computer use or external services can change data.

What developers need to change

New projects should use the Responses API and set the model to gpt-6-astra. Chat Completions remains supported, but OpenAI's migration guidance says tool calling with Astra requires Responses.

When migrating, remove unsupported sampling parameters such as temperature, top_p and top_logprobs. Applications previously using none or minimal reasoning should begin with low and compare quality, latency and cost on their own evaluation set.

The model is more sensitive to instructions in context, including skills and repository files such as AGENTS.md. Teams should audit those instructions, define when the agent may act autonomously and specify the desired writing and testing style.

Should you use GPT-6 Astra?

Astra is the strongest fit when the value of a correct, completed workflow matters more than the lowest token price: difficult coding, long research projects, computer-use agents and professional tasks that require several tools. Short classifications, simple summaries and high-volume routine requests may be more economical on a smaller model.

The practical way to decide is to test Astra and a lower-cost alternative on the same representative tasks. Measure completion quality, human correction time, latency and total API cost rather than comparing token prices alone.

Official sources

This article is based on OpenAI's GPT-6 Astra model page, model guidance and current model overview. Product access, prices and limits can change; check those pages before making a purchasing or deployment decision.