GPT-6 Astra

OpenAI's next-generation flagship for complex work, agentic execution and multimodal reasoning.

Context2M

ModalityText · Image · Audio

AccessAPI

ReasoningAdaptive

Where it fits.

Positioned as OpenAI's most capable work model; designed for long-horizon tasks, computer use, software engineering and enterprise workflows with stronger alignment safeguards. OpenAI positions GPT-6 Astra as its next-generation work model, released on September 9, 2026, with a 2M-token context window and native multimodal reasoning. It is designed for long-horizon agentic execution, computer use, software engineering and enterprise workflows.

Core capabilities

2M-token context window for long documents and sessions
Native multimodal reasoning across text, image and audio
Agentic computer use and software engineering workflows
Enhanced alignment and safety auditing for enterprise deployment

Best-fit workflows

Long-horizon agents and computer use
Complex software engineering
Enterprise knowledge and multimodal workflows

04 / OPERATIONS

From evaluation
to production.

Deployment notes

Start with low-permission evals on representative agent tasks; pin a stable model snapshot once available, enforce tool allowlists, and audit high-permission actions before broad rollout.

Watch before choosing

Rolling availability; verify access tier, regional support and safety policies before relying on it for high-permission agent workflows. Access is rolling out to limited enterprise partners first; verify availability tier, regional support and safety policies before deploying high-permission agents.

Family & versions

GPT-6 AstraFlagship
GPT-5.6 SolPrevious flagship
GPT-5.6 TerraBalanced
GPT-5.6 LunaEfficient

Specifications and positioning checked against provider documentation.