Home / Models / GPT-6 Astra

GPT-6 Astra

OpenAI · Multimodal model

GPT-6 Astra is OpenAI's flagship model, launched in September 2026 for demanding reasoning, coding, computer use and scientific work. It accepts text and images, has a context window of about 1 million tokens, and is the priciest model in OpenAI's…

Current
ReleasedSep 3, 2026
Context window1.05M tokens (922K max input, 128K max output)
Input → outputtext, image → text
API price$10 / 1M input tokens, $1 / 1M cached input tokens, $50 / 1M output tokens
Consumer accessChatGPT Work and Codex: Pro, Enterprise and Business Premium at launch; rolled out to Plus and Business over the following days (Enterprise admins must enable it)

Best for

The hardest multi-step reasoning, coding, computer-use and science work where cost per token matters less than capability.

Overview

GPT-6 Astra is OpenAI's flagship model, launched in September 2026 for demanding reasoning, coding, computer use and scientific work. It accepts text and images, has a context window of about 1 million tokens, and is the priciest model in OpenAI's GPT-6 lineup.

Strengths and limits

  • OpenAI's most capable model; claims state-of-the-art results on several agentic and science benchmarks
  • Operates apps without APIs (computer use) inside ChatGPT Work and Codex
  • 1M+ token context with low-to-max reasoning effort levels and Fast/Ultrafast modes
  • Most expensive OpenAI API model; no 'none' reasoning effort
  • Rated Critical for cybersecurity capability under OpenAI's Preparedness Framework, so it ships with extra safeguards and monitoring

Published benchmark scores

Terminal-Bench 4.057.9% Source: www.anthropic.com
OSWorld 2.0 (offline set, max effort)73.5% Source: community.openai.com
HealthBench Professional64.7 Source: deploymentsafety.openai.com
Humanity's Last Exam (with tools)57.2% Source: www.anthropic.com

Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →