GPT-6 Astra is OpenAI’s new high-end model for long-running agent work, coding, computer use and large-context applications. Its API model ID is gpt-6-astra, with a 1,050,000-token context window and up to 128,000 output tokens.
The important availability caveat is that Astra is rolling out, not generally available everywhere yet. OpenAI says enterprises in its Trusted Access Program receive access first, with API and ChatGPT Plus, Pro, Business and Enterprise availability following in the coming days. Check the model list for your project before designing a production migration around it.
This article is based on OpenAI’s official model page, current-model guide and model comparison. We have not independently benchmarked Astra.
GPT-6 Astra specifications
| Specification | GPT-6 Astra |
|---|---|
| Model ID | gpt-6-astra |
| Rollout | Trusted Access enterprises first; broader API and ChatGPT rollout follows |
| Context | 1,050,000 tokens |
| Maximum output | 128,000 tokens |
| Knowledge cutoff | April 30, 2026 |
| Input | Text and images |
| Output | Text |
| Reasoning | Low, medium, high, xhigh and max |
| Main API surfaces | Responses, Chat Completions and Batch |
| Structured outputs | Supported |
| Function calling | Supported |
OpenAI’s current guide recommends the Responses API for tool use. A model appearing in documentation does not guarantee that every account, region or endpoint has received rollout access.
GPT-6 Astra API pricing
Prices below are per one million tokens.
| Processing mode | Input | Cached input | Output |
|---|---|---|---|
| Standard | $10.00 | $1.00 | $50.00 |
| Batch | $5.00 | $0.50 | $25.00 |
| Flex | $5.00 | $0.50 | $25.00 |
| Fast | $20.00 | $2.00 | $100.00 |
OpenAI also lists cache writes at $12.50 per million tokens under standard processing. Batch and Flex are cheaper but serve different operational needs: Batch is asynchronous, while Flex trades availability and latency guarantees for price. Fast is the premium low-latency path; OpenAI says Fast does not support EU data residency.
Long-context pricing
When a request exceeds 272,000 input tokens, OpenAI applies the long-context rate to the complete request: 2× input and cached-input pricing, and 1.5× output pricing. Under Standard processing that means $20/M input, $2/M cached input and $75/M output.
Do not estimate a 600K-token request by charging only the tokens above 272K at the higher rate. The threshold changes the rate for the full request.
Built-in tools and agent capabilities
Astra supports OpenAI’s hosted agent tools on the Responses API, including:
- web search and file search;
- code interpreter and hosted shell;
- computer use;
- image generation;
- MCP and tool search;
- skills and apply patch;
- custom function calling and structured outputs.
This makes Astra a candidate for long-running engineering agents that need to inspect files, execute code, browse, operate a UI and call external tools in one workflow. It does not remove the need for permission boundaries, approval gates or output validation. See our AI application security guide and AI testing and evaluation hub before giving an agent consequential access.
Reasoning levels
The supported reasoning settings are low, medium, high, xhigh and max. Astra does not expose none or minimal on its model page. Higher effort can improve difficult multi-step work, but it can also increase latency and output-token cost.
Use workload-specific evaluations rather than setting max globally. Routine classification, extraction and deterministic tool routing may not justify the most expensive reasoning path.
Astra versus Sol, Terra and Luna
| Model | Standard input/output | Context | Best fit |
|---|---|---|---|
| GPT-6 Astra | $10 / $50 | 1.05M | Highest-complexity agents, computer use, long workflows |
| GPT-5.6 Sol | $5 / $30 list price | 1.05M | Frontier coding and agents at a lower list price |
| GPT-5.6 Terra | $2 / $12 | 1.05M | Balanced production workloads |
| GPT-5.6 Luna | $0.20 / $1.20 | 1.05M | High-volume, cost-sensitive work |
Sol currently also has a temporary qualifying-usage promotion. Compare effective account pricing, not only list prices. Our GPT-6 Astra vs GPT-5.6 Sol comparison covers the direct hierarchy decision.
Legacy Chat model note
OpenAI’s current model catalogue marks GPT-5.3 Chat and GPT-5.2 Chat as deprecated, and the dedicated chatgpt-4o-latest page says that API model has been removed. Those Chat aliases should not be confused with every similarly named API model. Confirm the exact model ID before migrating.
Should you migrate now?
Evaluate Astra now if your account has access and your workload genuinely benefits from its agent toolset, 128K output ceiling or highest reasoning tier. Keep a production fallback while rollout remains staged.
For most high-volume applications, Terra or Luna can be dramatically cheaper. For demanding coding agents, compare Astra with Sol on successful-task cost, latency, tool-call accuracy and human review time—not vendor benchmark headlines alone.
Verdict
GPT-6 Astra is OpenAI’s most capable documented agent model, but it is also expensive and still rolling out. Its value is clearest in complex, tool-rich workflows where fewer failed attempts can offset $10/M input and $50/M output. Do not label it universally available until OpenAI changes the rollout language.