A model guide for the GPT-6 family

OpenAI’s GPT‑6 family provides multiple models tailored for different tasks, from prototyping to multi-step workflows across codebases and APIs. The guide outlines how to match a model to specific workloads by balancing capability, cost, and latency. It emphasizes adjusting prompts and skills to ensure consistency in task execution and boundaries for model autonomy. Developers are advised to plan for long-running work using steering, async tools, and delegation to maintain workflow integrity.

For production deployment, OpenAI recommends using caching and compaction to manage context and costs. Cached input tokens can reduce expenses by up to 95% compared to uncached tokens, depending on the model. The guide also advises reusing shared context through prompt caching for recurring tasks and maintaining stable instructions and tool definitions. Monitoring task success, latency, and cost per successful task is critical before deployment.

The suite includes GPT‑6 Astra for complex reasoning, GPT‑6.1 Sol for coding and research, and GPT‑6 Luna for scalable, focused tasks like data extraction. Reasoning effort levels in the API allow users to adjust the depth of analysis, from routine tasks to high-effort debugging. Users can modify reasoning effort mid-conversation without disrupting cache, enabling flexibility in task execution.

OpenAI provides tools like the caching dashboard and diagnostics guide to track context reuse and identify inefficiencies. The deployment checklist highlights best practices, including measuring task success and latency, and reviewing data controls. The guide frames model choice and reasoning effort as an intelligence-to-price tradeoff, encouraging users to evaluate pricing and performance for their specific needs.