◈ Intermediate

GPT-6 Astra in Practice

Master GPT-6 Astra from the ChatGPT interface through the Responses API — computer use, 1M-token context, reasoning tiers, built-in tools, and production agent patterns.

📚 15 lessons 🕐 219 min total ✓ Free

What's New in GPT-6 Astra

Understand what's genuinely new in GPT-6 Astra — recurrent depth reasoning, key benchmarks, plan tiers, and how this model compares to GPT-5.6. Sets the stage for the split track ahead.

Reasoning Tiers: Thinking Harder on Demand

Master GPT-6 Astra's five reasoning tiers from low to max. Learn when to request more thinking, what recurrent depth delivers, and how to avoid overpaying for capability you don't need.

The 1M Token Context Window in Practice

Put the 1,050,000-token context window to work. Discover what it can hold, use cases that were impossible before, and how to structure long-context requests for best results.

Computer Use: Let Astra Work Your Screen

Enable computer use in ChatGPT and understand the screenshot-to-action loop that powers it. Covers task categories, safety constraints, and what to expect when things go sideways.

Computer Use Workflows: Real Automation Recipes

Five annotated walkthroughs — form filling, CRM updates, calendar management, web research, and frontend QA — with prompt templates for each and guidance on handling partial failures.

Deep Research and Long-Document Analysis

Use Astra's 1M context and reasoning together for multi-source research, scientific data analysis, and cross-document synthesis. Includes prompt patterns for traceable citations.

Voice, Vision, and Multimodal Astra

Use Astra through ChatGPT Voice (added Sep 9, 2026), image input, and tight vision-text-audio coupling. Covers when multimodal input earns its overhead versus when plain text is better.

Astra for Coding and Software Engineering

Explore Astra's coding superpowers — 74.1% on DeepSWE, 88% on SRE-Bench — and set up practical workflows for large-codebase review, bug triage, and defensive security analysis.

The Responses API: Foundation for Serious Work

Learn why the Responses API is now the default for complex work, how to structure your first call, and how to migrate from Chat Completions. Covers model ID, streaming, and access tiers.

Built-in Tools: Web Search, File Search, Code Interpreter

Survey all eight Responses API hosted tools and go deep on web search, file search with vector stores, and code interpreter — giving your app live data, RAG, and Python execution without extra infrastructure.

Reasoning Effort as a Parameter

Control Astra's thinking time with the reasoning_effort parameter (low → max). Covers cost and token tradeoffs per tier, persisted reasoning across turns, and how to measure the right setting.

Function Calling, Async Tools, and Hosted Shell

Define custom functions, use async tool calling to run tools in parallel, and leverage hosted shell and apply patch for server-side code execution — Astra's SRE-level capabilities in your app.

Structured Outputs: Guaranteed JSON

Guarantee well-formed JSON from every Astra response using response_format with a JSON schema. Covers schema design patterns, client-side validation, and when to use structured outputs versus function calling.

Prompt Caching and Cost Optimization

Design prompts for maximum cache hits, calculate the real cost difference between reasoning tiers, and decide when a cheaper model is the right call. Includes a simple cost-per-task logging pattern.

Capstone: Build a Production Astra Workflow

Put it all together: build a complete computer use workflow (Path A) or a Responses API agent using functions, file search, and structured outputs (Path B). End-to-end from setup to cost measurement.

Start Lesson 1 →