OpenAI has published a model guide for its GPT-6 family, offering practical tips for getting the best results from the models while managing time and cost.

The guide describes GPT-6 as "our most advanced suite of models yet," offering a choice of models for different kinds of work. It covers choosing a model, giving effective instructions, managing long-running work, and preparing for production.

On production use, OpenAI recommends using caching and compaction to manage context and cost, measuring task success and latency, and planning for monitoring and data controls. The company said cached input tokens cost up to 95% less than uncached input tokens, depending on the model.

The guide advises matching the model to the workload as an "intelligence/price tradeoff." GPT-6 Astra is positioned for the hardest reasoning work, GPT-6.1 Sol for complex coding, research, and computer use, and GPT-6 Luna for focused tasks at scale and everyday, repeated work with a clear goal.

Reasoning effort can be set in the API from low for routine tasks to extra high or max for difficult work, and can be changed mid-conversation without breaking cache. Fast mode and Ultrafast are offered when response time matters, with Ultrafast available for GPT-6 Astra.

"Models have gotten much better at understanding nuance and ambiguity, so overly specific guidance can now hinder results where it previously helped," said Eric Provencher, Developer Experience at OpenAI.

For long-running tasks, the guide points to steering, asynchronous tools, and delegation. Mid-turn steering lets users send corrections while the model works, asynchronous tool calling lets the model continue independent work while a slower task runs, and GPT-6.1 Sol supports multi-agent workflows in the Responses API, currently in beta.

Computer use lets GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna interact directly with websites and desktop apps, including applications without an API. The guide cites teams including Harvey, Cognition, Hex, and Invideo building with GPT-6 Astra, with Invideo reporting roughly three times the success rate on color-grading and correction tasks.

Full Article: A model guide for the GPT-6 family