Luna’s much lower token cost and new key controls create a practical way for small businesses to route routine AI work cheaply and reserve Sol for harder tasks.
Why the Sol-Luna split matters for small businesses
OpenAI’s new GPT-6 Sol and GPT-6 Luna models are less interesting as a product launch than as a pricing decision for businesses that already use AI in operations. According to OpenAI’s changelog, both models accept text and image inputs and generate text through the Responses API and Chat Completions API. The verified pricing brief says Luna is 20 times cheaper than Sol on both input and output in the short-context tier, and the same 20-times gap holds in the long-context tier. That creates a clear operating choice: use Luna for high-volume, lower-risk work, and reserve Sol for tasks where better reasoning is worth the higher cost.
For small businesses, that split is most useful where AI is already acting like a labor substitute. Routine customer support triage, lead qualification, FAQ chat, document parsing, OCR-plus-summary from image intake, and back-office workflow automation are all examples in the research brief where many requests are repetitive and can be handled at lower cost. In those cases, the business value is not the model name itself; it is the ability to reduce per-task AI spend enough to make automation practical at scale.
Where Luna can replace labor, and where Sol still earns its keep
The economics suggest a simple routing strategy. Luna is the better fit for first-pass work: classifying inbound messages, extracting fields from forms or photos, summarizing documents, answering common questions, and drafting standard replies. Those are the kinds of tasks where speed and low unit cost matter more than deep reasoning. If a workflow processes large volumes of similar requests, a cheaper model can lower operating costs without changing the customer-facing process.
Sol is the higher-capability option for harder cases that justify more spend. The brief frames Sol as the choice for “harder tasks,” while Luna is the economical option for routine work. That means businesses can design a two-step system: let Luna handle the bulk of requests, then escalate only the ambiguous, sensitive, or high-value cases to Sol. For service businesses, that can mean faster response times without hiring for every incremental workload spike. For consultants and agencies, it can mean offering AI-assisted intake or document review as a lower-cost service line while keeping expert review for the final pass.
New key controls reduce operational risk
OpenAI also added API key governance controls on Sep. 15, 2026, and key expiration controls on Sep. 10, 2026, according to the changelog. Administrators can restrict new key creation to service-account keys, allow only user-owned project keys, or disable new key creation entirely. The verified brief adds that organization-level restrictions override project settings and that existing keys are not affected. OpenAI also lets admins set expiration dates when creating project API keys and enforce a maximum key lifetime at the organization or project level.
For small businesses using contractors, agencies, or internal staff to build AI workflows, those controls matter because they reduce the chance of long-lived, unmanaged keys sitting in old projects or employee accounts. They also make least-privilege access easier to enforce. In practical terms, that lowers the risk of surprise usage, unauthorized deployments, and cleanup headaches when a freelancer leaves or a project changes hands. For owners who have delayed automation because of security concerns, these controls remove one of the common operational objections.
What to do now
The most practical move is to map AI work by risk and volume. Put routine, high-volume tasks on Luna first. Keep Sol for escalations, complex reasoning, and work where a mistake is expensive. If your business handles image-based intake, document-heavy workflows, or customer support, the new pricing tiers make it easier to justify an AI layer that actually saves time instead of becoming another overhead line.
Teams that operate in regulated or compliance-sensitive environments should also check the pricing and processing details before choosing a deployment path. The research brief notes that EU data residency for GPT-6 Sol and Luna is available only with Standard processing, and regional-processing endpoints carry a 10% uplift where eligible. That means the cheapest model is not always the cheapest deployment once residency or regional processing is required. For many small businesses, though, the bigger story is simpler: Luna makes routine automation cheap enough to use broadly, while Sol remains the premium option for the work that needs more judgment.
