Two tiers under the flagship
OpenAI released GPT-6 Sol and GPT-6 Luna on Tuesday, filling in the range below GPT-6 Astra, the frontier model it shipped on 3 September. Sol is pitched at complex work such as coding; Luna at clerical and high-volume tasks.
The company’s own framing is about cost rather than capability. “GPT-6 Astra introduced a new generation of intelligence,” OpenAI said; “these models extend its benefits by making that intelligence more efficient and accessible.”
What they cost
OpenAI’s published API rates put GPT-6 Sol at $2 per million input tokens and $10 per million output tokens. GPT-6 Luna is listed at $0.10 and $0.50. GPT-6 Astra, the flagship, remains at $10 and $50.
That makes Luna a hundredth of Astra’s input price. Batch and flex processing halve both figures again; a fast-mode endpoint doubles them. OpenAI told TechCrunch the new pair cost roughly 50 per cent less than the equivalent models in the GPT-5.6 generation they replace.
The accuracy claim
OpenAI says GPT-6 Sol makes about half as many mistakes as its predecessor and reaches Astra-level reliability at a much lower cost. That figure comes from the company’s internal factuality evaluations, not from an outside evaluator, and should be read as a vendor’s account of its own product until someone else measures it.
OpenAI also says the two models handle tasks substantially better than Anthropic’s Fable and Opus models. That comparison aged within hours: Anthropic released Claude Opus 5.5 the same day and cut its own prices by about 40 per cent, so the Opus figures OpenAI benchmarked against are no longer the current ones.
Where they show up
Sol and Luna are available in ChatGPT Work, in Codex and in the API for paid accounts. Luna also reaches the desktop app and Free and Go users, which makes it the model most people will actually encounter. The rollout across the ChatGPT app and website ran through Tuesday.
The placement matters more than it looks. Putting Luna in the free tier means OpenAI is serving its largest audience from the cheapest model in the generation, and reserving Astra for the paying edge of the customer base.
What this says about the market
Two frontier labs cut prices on the same day, in the same direction, at the same part of the range. Neither release is primarily about a new capability ceiling. Both are about the cost of running the capability that already exists — which is what a market looks like when the bottleneck moves from what a model can do to what it costs to let it do it at volume.
That pressure is coming from customers. Startups have spent the year moving workloads onto open-weight models to escape per-token bills, and enterprise agents multiply token use in ways that subscriptions were not priced for.
What to watch
Whether independent evaluators put Sol where OpenAI puts it, and whether the price of Astra moves. A flagship holding at $10 per million input tokens while the tier below it sells at $2 is a gap that either narrows or starts to look like a price for exclusivity rather than for capability.
