OpenAI adds GPT-6 Sol and Luna with lower API prices

OpenAI announced GPT-6 Sol and GPT-6 Luna as additions to the GPT-6 model family.[1] The company says the models were trained using methods similar to those used for GPT-6 Astra and bring Astra’s advances in professional work, factuality, coding, computer use and alignment to faster, more affordable models.[1] OpenAI says Astra remains its best model overall, so Sol and Luna are additions rather than replacements for its top model.[1] The developer community announcement names the API, Codex and ChatGPT as places where the models are being announced, without the evidence specifying further availability details.[2]

OpenAI cut Sol and Luna API prices by 50% compared with their GPT-5.6 promotional prices.[1] Sol costs $2 per million input tokens, down from $4, and $10 per million output tokens, down from $20.[1] Luna costs $0.10 per million input tokens, down from $0.20, and $0.50 per million output tokens, down from $1.20.[1] These listed prices are per million tokens.[1] OpenAI also says improved prompt caching raises cache hit rates by default and gives a 90% discount on cached input-token reads.[1] For API budgeting, compare expected input and output volumes separately and account for caching rather than relying on a single per-token figure.

OpenAI reports that Sol at xhigh effort scored 33.2% at $0.27 per task, outperforming Claude Opus 5 at max effort for 9% of Opus 5’s per-task cost.[1] It says Sol at max effort scored 68.8%, within 1.1 percentage points of Claude Fable 5’s highest score, at approximately 80% lower cost per task.[1] OpenAI says Luna at max effort scored 66.6%, comparable to Opus 5 and Fable 5 at medium effort, while costing 93% less than Opus 5 and 96% less than Fable 5 per task.[1] These are company-reported comparisons at specified effort levels, not a single universal ranking.

Artificial Analysis reported that Sol and Luna’s Intelligence Index and Coding Agent Index scores remained level with GPT-5.6 overall, with mixed results across individual evaluations.[3] Its reported Coding Agent Index score for Sol was 57, two points above GPT-5.6 Sol.[3] It also reported that Sol’s AA-Omniscience hallucination rate fell from 92% to 60%, while Luna’s fell from 93% to 77%.[3] OpenAI separately says Sol makes about half as many mistakes as its predecessor, and that Luna matches GPT-5.6 Sol at higher effort levels at about one-hundredth the cost.[1] The reported results vary by model and evaluation, so teams should test against their own tasks.