GPT 6.2 Astra by OpenAI
OpenAI GPT 6.2 Astra is a long-context reasoning model that accepts text and images. Use it for coding, document work, and tool-driven agents.
## Overview
GPT 6.2 Astra is OpenAI's most capable reasoning model, built for long, complex work. It takes text plus optional images and returns text you can act on. It can search the web and call the tools you define, which makes it a strong fit for agents that work across large codebases, long documents and multi-stage plans with fewer back-and-forth steps.
## What you can build
- Codebase agents that plan, edit and summarize large refactors
- Visual QA assistants that answer questions about screenshots, UI states or diagrams
- Long-document reviewers that extract requirements, risks and action items from large specs
- Tool-using copilots that search the web, call your functions and report results
- Multi-step workflows that continue across turns with tool results
## Inputs
- A written request that describes your goal, constraints and desired output format.
- An optional multi-turn conversation history (messages) when you need continuity across turns.
- Optional images for visual reasoning: PNG, JPEG, WEBP or non-animated GIF, up to 10 per request.
- Optional system instructions that set rules and style for the run, up to 5,000 characters.
- Optional user and session IDs to keep chat history tied to a user.
- A reasoning effort level: low, medium, high, xhigh or max. Astra doesn't accept runs with reasoning turned off.
- An optional web search toggle for information newer than the model's knowledge cutoff. Each search costs $0.01.
- A verbosity preference that controls how short or detailed the answer is.
- An optional maximum output length, up to 128,000 tokens.
- Optional tool definitions, described as functions with JSON inputs, when you want the model to call your tools.
- Optional tool choice: let the model decide, forbid tools, or require a tool call.
- For multi-step tool workflows, continue from a prior run by passing its task token and the tool results.
## Outputs
- A text answer.
- Tool-call requests when tools are enabled, each with a tool name and a JSON input.
## Recommended settings
- Medium reasoning effort for most production workloads.
- Low reasoning effort for short classification, rewriting or fast responses.
- High or xhigh reasoning effort for deep debugging, multi-step planning or tricky logic where mistakes are costly.
## Limitations
- Outputs text only. Audio and video input are not supported.
- Image understanding depends on input quality. Blurry screenshots, tiny text or heavy compression reduce accuracy.
- The knowledge cutoff is May 18, 2026. Turn on web search for newer facts.
- Requests over 272K input tokens are billed at higher long-context rates.
## Safety & compliance
- OpenAI applies safety filtering, and requests for harmful assistance such as unauthorized intrusion, weaponization or dangerous biological work are refused.
- Use the model only for lawful purposes.
- If you handle personal or sensitive data, minimize what you send and follow your organization's privacy requirements.
API quick start
Run gpt 6.2 astra with a single API call.
{
"prompt": "How to roast coffee beans at home?",
"inputImage": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}curl -X POST "https://api.wiro.ai/v1/Run/openai/gpt-6-2-astra" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "How to roast coffee beans at home?",
"inputImage": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}
JSONOther models you might like
Explore similar llm & chat models.



