OpenAI launches GPT-6 model family with three tiers for varying reasoning needs and costs

OpenAI has released the GPT-6 model family, introducing three new variants designed to give organizations finer control over the tradeoff between reasoning capability, speed, and cost. The lineup includes GPT-6 Astra for maximum intelligence, GPT-6.1 Sol for a balance of performance and price, and GPT-6 Luna for high-volume, cost-sensitive tasks. The release also brings new API features including asynchronous tool calling, mid-turn steering, and the ability to change reasoning effort mid-conversation.

Three models for different workloads

GPT-6 Astra is the most capable model in the new family. It achieved stronger evaluation results using substantially fewer output tokens than earlier models, and OpenAI said its estimated API cost per task was lower despite higher per-token pricing. Astra excels at multi-step workflows across code, browsers, and professional software. It is also designed to exercise more care around task boundaries, asking focused clarifying questions when additional input could change the outcome rather than making assumptions.

GPT-6.1 Sol offers what OpenAI describes as near-Astra performance at a lower cost for complex coding, computer use, and professional work. GPT-6 Luna is positioned as the fastest and most cost-effective option for focused, high-volume tasks. Both GPT-6 Astra and GPT-6.1 Sol do not support the "none" reasoning effort setting; users should use "low" instead. GPT-6 Sol and GPT-6 Luna do support "none."

New API capabilities

The release introduces several features aimed at improving workflow efficiency. Async tool calling lets the model continue reasoning or call other tools while an application runs a function in parallel, returning results when ready using the original call ID. Mid-turn steering allows developers to send additional instructions-such as corrections or requirement changes-while the model is working, with the API preserving completed work and incorporating the update.

Another addition is the ability to change reasoning effort mid-conversation without breaking the prompt cache. Using a configuration update input item, developers can increase reasoning effort for difficult segments of work or reduce it for routine follow-ups. OpenAI has also added misalignment monitoring for GPT-6 Astra, which asynchronously checks for potential issues and triggers alerts when necessary.

Prompting for autonomy and clarity

OpenAI provided detailed prompting guidance for GPT-6 Astra, which tends to ask more clarifying questions than previous models. For teams that want more autonomous execution, the company recommends prompts that explicitly instruct the model to infer intent, bias toward action, and persist until a task is complete. One suggested prompt reads: "You should infer the user's intent and task scope from the instructions and prior conversation context. Your job is to bias towards action and carry the user's intended task to completion."

The model is also more sensitive to instructions in skill files and other context, which means unclear guidance can cause it to pause or block work early. OpenAI recommends auditing files like AGENTS.md and making the priority of user instructions explicit. For writing style, GPT-6 Astra defaults to detailed, formatted responses with lists and tables. Teams that prefer prose with less formatting should specify that preference directly in prompts.

Why this matters for executives, writers, and communications professionals

The GPT-6 family changes how organizations should think about model selection. Instead of defaulting to the most capable model, teams can now match specific tasks to the right tier-reserving Astra for work where quality directly affects business outcomes, using Sol for complex but cost-conscious projects, and routing routine high-volume tasks to Luna. For writers and communications professionals, the model's tendency toward formatted, list-heavy responses means prompt engineering around tone and structure becomes more important, not less. The new mid-turn steering feature also means editors can course-correct outputs without restarting a session, which may reduce iteration time on drafts. Professionals looking to build these prompting and model-selection skills can explore OpenAI API Courses or AI for Developers Courses to develop practical workflows with the new model family.