GPT-6 Astra Launches: An AI for Sustained, Complex Work
OpenAI unveiled GPT-6 Astra on September 3. It can carry out coding, research, and computer tasks in a single workflow and accept additional instructions to change direction while a task is underway. Official model information
Changing Direction During a Task
Two notable features in the developer documentation are asynchronous tool calls and additional instructions during a task. The model can continue independent work while a tool is running. When requirements change, it can incorporate follow-up instructions while preserving completed work. App developers must connect tool execution and result delivery themselves. Official feature guide
For example, imagine asking, midway through creating a webpage, "Skip deployment and stop after checking it locally." The ability to redirect a task this way is useful during lengthy development work. However, the exact support depends on the app or API integration being used.
A Model for Long Materials
The official specifications list a context limit of 1,050,000 tokens and a maximum output of 128,000 tokens. The API model ID is gpt-6-astra. It also supports tools such as web search, file search, and computer use. Being able to provide large amounts of material is different from processing every detail accurately without omissions. Model specifications
Check Input Length When Calculating API Costs
Standard prices per million tokens:
| Item | Price |
|---|---|
| Input | $10 |
| Output | $50 |
| Cached input | $1 |
| Cache writes | $12.50 |
<p class="blog-price-source"><a href="https://developers.openai.com/api/docs/models/gpt-6-astra" data-i18n="blog.price.source">Official pricing</a><span data-i18n="blog.price.checked"> · Checked September 5, 2026</span></p>
If the input exceeds 272,000 tokens, input and cache charges are doubled and output charges are multiplied by 1.5 for the entire request. Tool usage may incur additional charges too, making it difficult to estimate the total cost from base rates alone. Official pricing conditions
A Launch Does Not Mean Your Account Has Access
A model's release does not mean every account can use it at the same time. Check the service's model selection screen, organization permissions, and API access. Do not interpret the API prices in this article as ChatGPT subscription fees.
If you are considering it for development, try giving it a task you have already completed under the same conditions. Comparing accuracy, the time spent making corrections, and actual cost can help determine whether it fits your work. This is a usage suggestion from GamZip, not a performance review based on our own measurements.