OpenAI published a GPT-6 family guide on October 2 covering model selection, prompting and production workflows. It recommends matching capability, reasoning effort and processing speed to the workload instead of choosing one configuration for every task.
The guide discusses prompt caching, context compaction and measuring success, latency and cost before deployment. For long-running work, it covers steering, asynchronous tools and delegation, with clear boundaries for user input. These are implementation recommendations, not a guarantee of task performance. Official guide.