OpenAI calls Astra its most capable broadly deployed model—built to take difficult work from request to result.

It can plan and complete multistep workflows instead of waiting for instructions at every small step.

OpenAI says Astra can navigate software, research online, fill forms and troubleshoot what it sees on screen.

The goal is end-to-end engineering: understand a project, edit code, use tools, test changes and keep going.

Astra is trained to produce polished work that follows templates, instructions and changing requirements.

OpenAI highlights browsing, scientific analysis and research as major areas of improvement for Astra.

The API model page lists a 1,050,000-token context window for extensive codebases and source material.

Astra supports up to 128,000 output tokens—enough for lengthy code, analysis and professional documents.

OpenAI says Astra is its first broadly deployed model to reach the Critical cybersecurity threshold.

Standard pricing per million tokens is $10 input and $50 output, with separate cache rates and conditions.

Astra is rolling out across eligible ChatGPT plans and the API. Timing can vary by account and workspace.