OpenAI expanded its GPT-6 line on Tuesday with Sol and Luna, two models tuned for work that repeats thousands of times a day. Both sit below Astra, the flagship the company shipped earlier this month.
Pricing is the headline. API rates for the pair land at about 50 percent of what GPT-5.6 Sol and Luna charged during their promotional window. Luna’s output rate alone drops from $1.20 per million tokens to $0.50, a cut closer to 58 percent. OpenAI credits better caching and inference for the saving.
The tiers split by task. Sol takes harder coding and professional jobs, while Luna handles summarising, extraction and quick answers at volume. MarkTechPost reports Sol scoring 68.8 percent on DeepSWE v1.1 at max effort, 1.1 points behind Claude Fable 5 at roughly 80 percent lower cost per task.
On AutomationBench 1.0.6, Sol reaches 33.2 percent at xhigh effort for $0.27 a task, against 26.9 percent for Claude Opus 5 at 11.1 times the cost. OpenAI says Sol makes about half as many factual mistakes as its predecessor on an internal test built from flagged user conversations.
Both models are API-only, so there are no weights to self-host, and both are already generally available on Amazon Bedrock. OpenAI also reworked prompt caching for GPT-6, where eligible shared prefixes reused inside 30 minutes earn reads discounted by up to 90 percent.
