GPT-6 Astra

https://openai.com/index/gpt-6-astra/

$10/M input tokens, $1/M cached input, $50/M output

Async Tool Calling — Continue useful work while another tool is running
Mid-Turn Steering — Add corrections or new requirements during execution
Professional Artifacts — Create documents, spreadsheets, presentations, sites, and apps
Computer Use — Operate browsers and professional software in multi-step workflows
Developer Tools — Hosted shell, apply patch, code interpreter, functions, and structured outputs
Agent Integrations — MCP, tool search, web search, file search, and multi-agent orchestration
Long Context — 1.05M-token context window with up to 128K output tokens
Adjustable Reasoning — Low, medium, high, xhigh, and max effort levels
Multimodal Input — Text and image input with text output
Safety Monitoring — Additional controls for agent behavior and critical cyber capability

Excellent end-to-end computer and browser use
Async tool calling reduces idle time in long workflows
Mid-turn steering lets users redirect active tasks
Creates polished documents, spreadsheets, presentations, sites, and apps
1.05M-token context window and 128K maximum output
Broad tool support including computer use, hosted shell, MCP, and code interpreter
Strong vendor-reported gains on AutomationBench, OSWorld 2.0, and BenchCAD

Premium API price: 2.5x GPT-5.6 Sol for standard input and output
Phased rollout means access is not yet universal
Not available on the free ChatGPT tier
Advanced cybersecurity tasks are restricted outside approved Daybreak access
Vendor-reported benchmarks still need validation on real workloads
Asynchronous safety monitoring may pause or stop some agent runs