Summary
On September 3, 2026, OpenAI released GPT-6 Astra, a frontier model tuned for coding, research, computer use, and long multi-step work, with a roughly 1.05M-token context, 128K max output, and text and image input. Priced at $10 per million input and $50 per million output tokens ($1 cached input), it adds asynchronous misalignment monitoring that can pause agent runs in supported Responses API requests.
What changed
OpenAI shipped GPT-6 Astra (API id gpt-6-astra) with a 1,050,000-token context, 128K output, text and image input, and an April 30, 2026 knowledge cutoff. It introduces misalignment monitoring that asynchronously checks agent work and can pause or stop a conversation for review. Enterprises in OpenAI's Daybreak cyber cohort got access September 3, with general availability September 4.
Why it matters
A frontier model built for long-horizon agents plus built-in misalignment monitoring signals that agent safety controls are moving into the model API itself, and its pricing and context window reset the reference point competitors like Anthropic and Google are measured against.
Evidence excerpt
OpenAI released GPT-6 Astra on September 3, 2026. 1,050,000-token context, 128K max output. Pricing is $10 per million input tokens and $50 per million output tokens, with cached input at $1. Misalignment monitoring asynchronously checks for potential issues during agent work in supported Responses API requests.