OpenAI publishes a model guide for the GPT-6 family

#What happened

OpenAI published "A model guide for the GPT-6 family" on October 2, 2026 — effectively the instruction manual for the GPT-6 lineup it has been shipping since early September. The guide sorts the family into three primary models: GPT-6 Astra as the flagship for the hardest reasoning workloads, GPT-6.1 Sol for complex coding, research, and computer-use tasks, and GPT-6 Luna for high-volume workloads with clear objectives such as extraction, classification, and structured summaries. Pricing: Astra $10 input / $50 output per million tokens, 6.1 Sol $2 / $10, Luna $0.10 / $0.50.

#Built for long-running agents

The guide leans hard into autonomous, long-running work. Models can now take on tasks spanning hours or days, supported by mid-run steering — developers can update instructions while a task runs without canceling existing tool calls — plus asynchronous tool calls and parallel work. GPT-6.1 Sol supports multi-agent workflows in beta through the Responses API: it can delegate independent portions of a task to subagents and combine their findings. GPT-6 Astra extends that approach in Codex, where it can ask users for clarification while continuing portions of a task that do not depend on the answer. OpenAI recommends explicitly defining which decisions an agent can make independently and which require approval.

#The cost advice

The guide pushes prompt caching and context compaction to control the cost of long agent workflows — cached input can cost up to 95% less than uncached tokens depending on the model. It also tells developers to simplify prompts: increasingly capable systems handle nuance and ambiguity well, and overly prescriptive instructions can sometimes hurt performance. All three models get computer-control capabilities to interact with websites and desktop apps when APIs are unavailable — OpenAI's example: an agent investigates a software bug, modifies the code, then opens the product in a browser to verify the fix.

#Why it matters

This is the first official routing guide for the family that now powers everything from ChatGPT to Codex to Dots. With 6.1 Sol landing one point behind Astra on independent benchmarks at a fifth of the cost (already reported), the guide's real message is operational: match the model to the job, steer agents mid-run instead of restarting them, and measure cost per accepted result — not per million tokens.

Verified October 2, 2026.

Sources: https://cryptobriefing.com/openai-gpt-6-family-model-guide/ (report), https://particle.news/story/openai-launches-dots-persistent-gpt6-agents-for-paid-users (context)