OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026. Both accept text and image inputs and generate text through the Responses and Chat Completions APIs. They sit alongside GPT-6 Astra, giving developers a choice among models aimed at different kinds of work.
What is GPT-6 Sol?
OpenAI describes GPT-6 Sol as a model for complex coding and agentic workflows. Its place in the lineup is between the highest-capability Astra and the more economical Luna: OpenAI's model catalog recommends Sol when teams want to balance intelligence and cost.
That positioning makes Sol a starting point for software agents that need to plan, use tools and complete demanding tasks. It does not guarantee that Sol will be the best or cheapest choice for every application; teams still need to measure task success, latency and total cost on their own workload.
What is GPT-6 Luna?
OpenAI positions GPT-6 Luna for focused, high-volume work. The model catalog recommends it when cost matters and requests are repeatable. In practice, that makes Luna a sensible candidate to evaluate for classification, extraction, short-form drafting and similar tasks with clear acceptance criteria. Those examples are applications inferred from its positioning, rather than claims that OpenAI has benchmarked those exact tasks for every user.
Luna supports reasoning effort settings from none through max, just like Sol. A lower price per token is useful, but the real decision depends on whether Luna reaches the quality threshold for the job.
GPT-6 Sol vs Luna: API pricing and specifications
For standard API requests with prompts of up to 272,000 input tokens, OpenAI lists the following September 22 launch prices. Prices are in US dollars per one million tokens.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6 Sol | $2.00 | $0.20 | $10.00 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
On those listed rates, Luna costs one twentieth as much as Sol per input or output token. That is a per-token comparison, not a promise of a 95% lower bill for a complete task: token use, tools and processing tier can change the total. OpenAI says prompts above 272,000 input tokens use higher rates for the full request, and other processing tiers have separate pricing.
Both Sol and Luna list a 1,050,000-token context window and a maximum output of 128,000 tokens. Each accepts image input but does not list audio or video input support on its model page.
What developers should check before switching
OpenAI's GPT-6 guidance recommends the Responses API for reasoning with tools. There is a specific Chat Completions constraint: Sol and Luna support function calling there only when reasoning_effort is set to none. Applications that combine reasoning and tool use should test the Responses API path.
For a first evaluation, run the same representative tasks on both models, record whether the outputs meet your acceptance criteria, and compare latency and total API cost. Sol is the stronger fit to test first for complicated coding and agent workflows; Luna is the economical fit to test for narrow work performed frequently. The right choice may also vary between stages of one workflow.
The broader picture
The September release broadens the GPT-6 family beyond Astra. For developers, the useful shift is choice: a model aimed at demanding workflows and another aimed at volume, with the same large stated context window. The official specifications are a starting point; real-world performance still has to be checked against the task at hand.