OpenAI launches cheaper GPT‑6 Sol and Luna

OpenAI has expanded its GPT‑6 line with Sol and Luna, mid- and low-cost models trained with methods similar to GPT‑6 Astra, the flagship the company introduced earlier this month. API pricing for both is 50% below their GPT‑5.6 promotional rates, which OpenAI attributes to caching and inference improvements.

The pitch is cost per task rather than raw capability. On AutomationBench, a business-workflow evaluation, Sol at xhigh effort outperforms Claude Opus 5 at max effort at 9% of Opus 5's cost per task; Luna at high effort gains 5.4 percentage points over its predecessor at 58% lower cost. On Agents' Last Exam, Sol at max effort scores 56.4%, which OpenAI says exceeds Opus 5's best result at 60% lower cost. On DeepSWE v1.1, Sol at max scores 68.8% against Claude Fable 5's 69.9% at xhigh, at roughly 80% lower cost per task; Luna scores 66.6%, at 93% below Opus 5 and 96% below Fable 5. On OSWorld 2.0 offline, Sol at xhigh hits 60.5% versus Opus 5 medium at 60.3%, again about 80% cheaper.

OpenAI also says Sol makes about half as many mistakes as its predecessor on an internal factuality eval built from de-identified user-flagged conversations, and that GPT‑6 prompt caching now yields higher default hit rates with a 90% discount on cached input reads, with reasoning-effort and tool-availability changes no longer invalidating cache.

The announcement references a price table that is not included in the material reviewed; specific per-million-token rates and availability dates are not stated.