GPT-6 Astra: Pricing, Context Window and Model Guide

What GPT-6 Astra is

OpenAI introduced GPT-6 Astra on September 3, 2026, as its first sixth-generation model and the new top of its lineup, taking over from GPT-5.6 Sol. OpenAI's framing was broad: state-of-the-art results claimed across computer use, browsing, software engineering, cybersecurity, science and professional work. Three weeks later OpenAI released GPT-6 Sol and GPT-6 Luna, trained with similar methods but built to be faster and cheaper, and then GPT-6.1 Sol, which OpenAI says approaches Astra on several evaluations. That sequence matters for anyone choosing a model, because Astra is now the expensive end of a family rather than the only member of it.

Who can use it

At launch OpenAI described a staged rollout that started with enterprises in its Trusted Access Program. Its API documentation now lists gpt-6-astra as available through both Chat Completions and Responses, with Batch supported and the standard five usage tiers, running from 500 requests per minute at Tier 1 to 15,000 at Tier 5. We read that as the API rollout having completed; if your account cannot see the model, the tier your organisation sits in is the first thing to check.

Pricing

Standard pricing is $10 per million input tokens, $1 per million cached input tokens and $50 per million output tokens. Batch and Flex run at half those rates, and Fast mode at twice. That is five times GPT-6 Sol on input and output, and a hundred times GPT-6 Luna. Our view is that the price makes Astra a model to route to rather than default to: put it on the tasks where the cheaper models measurably fail on your own data, and let Sol or Luna take the rest.

Context, output and reasoning

The context window is 1,050,000 tokens and maximum output is 128,000. The knowledge cutoff is April 30, 2026. Reasoning effort can be set to low, medium, high, xhigh or max; unlike Sol and Luna, Astra's documentation does not list a none setting, so every request reasons to some degree. Input is text and images, output is text.

Tools and endpoints

Astra supports the full set of OpenAI's built-in tools: web search, file search, image generation, the code interpreter, a hosted shell, apply_patch, skills, computer use, MCP servers and tool search. It runs on both Chat Completions and Responses. For agent work that leans on several of those tools at once, the Responses API is the one OpenAI builds its newer tool features around.

How it measures up

On LMArena's text board Astra has 1476 Elo (rank 37, max effort); in coding, 1542 Elo in Coding (rank 11, max effort). Epoch AI records 96% on GPQA Diamond. Those are independent measurements. The confidence range on that Elo is on our benchmarks page.

Official Resources