OpenAI's new top-of-line model, sitting above the whole GPT-5.6 family — a phased rollout that started with ChatGPT Pro, Business, and Enterprise, with Plus following within days.
GPT-6 Astra bills $10 / 1M input and $50 / 1M output for prompts up to 272K tokens; longer prompts are billed at $20 / $75 per 1M. Cached input reads are $1 / 1M (≤272K) or $2 / 1M (>272K), and cache writes are $12.50 / 1M.
OpenAI released GPT-6 Astra on September 3, 2026, calling it the most intelligent and aligned model it has shipped, setting a new state of the art on computer use, browsing, software engineering, cybersecurity, and professional work. Rollout was phased rather than immediate everywhere: ChatGPT Pro, Enterprise, and Business Premium subscribers got access first, inside ChatGPT Work and Codex, with Plus and Business users following "in the coming days" per OpenAI's own announcement. It's also available through the OpenAI API and on Amazon Web Services.
At $10 input / $50 output per 1M tokens (up to 272K tokens), Astra costs 2.5× the promotional rate of the prior flagship, GPT-5.6 Sol ($4/$20). Prompts beyond 272K tokens step up to $20/$75. The context window — 1.05M tokens with a 128K output ceiling — is the largest OpenAI has shipped.
Astra is positioned above the entire GPT-5.6 family for the hardest professional and agentic work: long-horizon software engineering, computer and browser use, and cybersecurity tasks where OpenAI says it sets a new frontier. As with prior flagship launches, expect the model picker and exact usage allowances on ChatGPT plans to keep changing during the rollout window — check OpenAI's help center for the current state on your plan.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| GPT-6 Astra | $10 | $50 | 1.05M |
| GPT-5.5 pro | $30 | $180 | 1M |
| GPT-5.6 Sol | $4 | $20 | 1M |
| Claude Fable 5.1 | $10 | $50 | 1M |
| Claude Opus 5 | $5 | $25 | 1M |
Astra prices exactly level with Claude Fable 5.1 at $10/$50, and well under GPT-5.5 pro's $30/$180. For work that doesn't need Astra's top-end reasoning, GPT-5.6 Sol remains the value flagship at a quarter of the input cost.