OpenAI has expanded the GPT-6 family with two lower-cost models, GPT-6 Sol and GPT-6 Luna. The company is positioning them for professional work, coding, automation and computer-use tasks, while keeping GPT-6 Astra as its top option for the most demanding workloads.
The clearest change is pricing. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna costs $0.10 and $0.50 respectively. OpenAI says those rates are 50% below the promotional pricing of their GPT-5.6 predecessors.
What happened
OpenAI says Sol and Luna were trained with methods similar to those used for GPT-6 Astra and inherit advances in reasoning, factuality, coding, computer use and alignment. The company attributes the lower prices partly to improvements in inference efficiency and prompt caching.
Prompt caching is a major part of the economics. OpenAI says cached input-token reads receive a 90% discount and that GPT-6 is better at reusing prompt prefixes by default. For long-running agents, that can reduce the cost of repeatedly processing the same codebase, instructions or conversation history.
The models are available in Codex and ChatGPT Work for Plus, Pro, Business, Enterprise and Edu users. GPT-6 Luna is also available to Free and Go users in the desktop app. In the API, developers can call gpt-6-sol and gpt-6-luna.
OpenAI reports substantial gains over GPT-5.6 Sol on coding and professional-work evaluations. Those benchmark results come from OpenAI’s own research setup, so they are useful directional evidence rather than a guarantee of performance for every production workload.
Why it matters
Agentic workloads multiply model costs because a single user request can trigger many tool calls, code edits and context passes. Lower token prices and cheaper cached context therefore matter more for agents than they do for short chat interactions.
The release also intensifies competition in the middle of the frontier-model market. Developers can run more iterations with stronger models before cost becomes the limiting factor, which can change which tasks are economical to automate.
OpenAI is also making its product segmentation clearer. Astra remains the premium tier, while Sol and Luna extend GPT-6 capabilities across lower price points rather than replacing the flagship outright.
