GPT-6 Astra

What OpenAI says. The Using GPT-6 guide calls Astra the "highest intelligence" option in the family, built for multi-step workflows across code, browsers and professional software. OpenAI also states that in several of its own evaluations Astra reached stronger results with substantially fewer output tokens, and that its estimated API cost per task was lower than earlier models despite the higher per-token price. That is a vendor claim, not something we have checked. The model selection guide says that if cost and latency are not a concern you can default to Astra, and maps effort levels to work: low for concise writing that preserves facts, medium for ambitious projects that need broad context, extra high for demanding analysis with exacting requirements.

Our editorial judgment. Astra makes most sense as the top rung of a routing setup, not as the default for every call. Use it where a failed task means human rework, such as a refactor across many files or a research memo built from conflicting sources. For chat or extraction at volume, the pricing gap to Sol and Luna needs evidence from your own tests. If you are comparing coding assistants rather than raw models, see our AI coding tools guide.

• Long-prompt surcharge. Prompts above 272K input tokens are billed at 2x input and cache rates and 1.5x output for the whole request. At list prices that moves input to USD 20 and output to USD 75 per 1M tokens for those requests.
• Cache writes cost extra. Writing to the cache is billed at USD 12.50 per 1M tokens, so caching pays off only when a prefix is reused.
• No none reasoning. The guide says to use low instead when migrating code that set none; requests that set minimal should start at low and be compared.
• Tool calling needs Responses. Astra supports Chat Completions, but the guide states tool calling requires the Responses API.
• Unsupported parameters. When reasoning effort is not none, remove temperature, top_p and top_logprobs.
• Data residency. Fast mode supports EU data residency for other GPT-6 models but not Astra. Astra Ultrafast supports US data residency and global processing only, and Fast mode for Astra has no latency SLA.
• Rate limits by tier. Build: 5,000 RPM, 1,000,000 TPM, 3,000,000 batch queue. Launch: 10,000 RPM, 4,000,000 TPM, 200,000,000 batch queue. Grow: 15,000 RPM, 40,000,000 TPM, 15,000,000,000 batch queue.
• Behavior notes. The guide warns Astra is more likely to stop and ask clarifying questions and can be sensitive to instructions in files such as AGENTS.md.

All three GPT-6 models share the same 1,050,000 context window, 922,000 maximum input, 128,000 max output, text and image input, and the same Responses API tool list. The differences are price, reasoning options and residency.

OpenAI positions Sol as "near-Astra performance at a lower cost" and tells developers to compare the two on their own tasks. Luna is described as the most efficient model for focused, high-volume work. Outside OpenAI, Grok 4.7 lists USD 2.00 input and USD 6.00 output per 1M tokens with a 500,000 token context, according to its vendor docs.

1. Pick 20 to 50 real tasks from your backlog where you already know what a good result looks like, and include a few that previous models failed.
2. Run each task on Astra at medium effort and on Sol at medium effort with identical prompts, tools and context, and record pass or fail against your own rubric.
3. Log input, cached, and output tokens per task and multiply by list prices to get cost per completed task, not cost per call.
4. Count clarifying questions and stalls separately, since OpenAI says Astra asks more often.
5. Keep Astra only where its pass rate gain covers the price difference.

We read the OpenAI model page, the Using GPT-6 guide and the model selection guide on 15 October 2026. We have not run GPT-6 Astra, and nothing on this page is a test result or ranking. Prices and limits change; check the vendor page before budgeting. See our ranking method for how HumanRank scores tools and the model directory for other profiles.

What is GPT-6 Astra? It is OpenAI's top GPT-6 model, described as its most capable model for complex reasoning, coding, computer use, research and document creation.

How much does GPT-6 Astra cost? It costs USD 10 per 1M input tokens and USD 50 per 1M output tokens at list price on 15 October 2026, with cached input at USD 1 per 1M tokens.

What is the GPT-6 Astra context window? It is 1,050,000 tokens, with a maximum of 922,000 input tokens and 128,000 output tokens.

Does GPT-6 Astra support batch processing? Yes, the Batch endpoint is supported and Batch and Flex are priced at 50% of Standard rates.

Can GPT-6 Astra process audio or generate images directly? No, it accepts text and images and outputs text only, though the Responses API lists an image_generation tool.

Is GPT-6 Astra worth it over GPT-6.1 Sol? Only if your own tests show a higher completion rate on your tasks, because Sol is listed at one fifth of Astra's token price and OpenAI itself says to compare them.