OpenAI Releases GPT-6 Sol and Luna: 50% Cheaper API Pricing
OpenAI launches GPT-6 Sol and Luna API-only models with 50% lower pricing, improved caching, and benchmarks outperforming competitors.
Stock photo for illustration only, not from the actual event
- OpenAI introduces two new GPT-6 models: Sol for professional tasks and Luna for high-volume everyday work.
- API pricing for both models is slashed by 50%, with Luna's output price dropping by 58%.
- Sol delivers competitive benchmark scores in coding and task automation at a fraction of competitors' costs.
- An upgraded prompt caching system offers discounts of up to 90% on cached input reads.
OpenAI has officially expanded its artificial intelligence portfolio by releasing two new models within the GPT-6 family: gpt-6-sol and gpt-6-luna. Both models are immediately accessible through the OpenAI API as API-only solutions, meaning there are no weights available for self-hosting.
The GPT-6 family now features three distinct tiers. Astra remains the premier model for the most demanding workloads, while Sol targets complex coding and professional tasks at a reduced cost. Meanwhile, Luna is engineered for fast, high-volume everyday operations.
Stock photo for illustration only, not from the actual event
According to the OpenAI team, enhanced caching and inference architectures enable them to serve these models more economically. Consequently, the company is cutting API pricing for Sol and Luna by 50% compared to their GPT.5.6 promotional rates. A notable detail is that Luna's output price drops from $1.20 to $0.50, representing a reduction of roughly 58%.
In professional evaluations, Sol scored 33.2% on AutomationBench 1.0.6 at an xhigh effort setting for $0.27 per task, whereas Claude Opus 5 scored 26.9% at 11.1 times that cost. On Agents’ Last Exam, Sol achieved 56.4% at max effort, outperforming Claude Opus 5's top score while operating at a 60% lower cost per task.
The introduction of the Sol and Luna tiers underscores OpenAI's aggressive commercial strategy to capture developers seeking an optimal balance between high-end performance and lower operational expenses. By coupling a 50% reduction in API pricing with benchmark results that rival major competitors like Anthropic, OpenAI is signaling that the AI race is increasingly driven by economic viability and large-scale practical efficiency rather than raw capability alone.
Regarding coding benchmarks, Sol achieved 68.8% on DeepSWE v1.1 at max effort—trailing Claude Fable 5 by just 1.1 points while reducing task costs by roughly 80%. Luna scored 66.6% at max effort, performing comparably to Opus 5 and Fable 5 at medium effort while running at 93% and 96% lower costs per task, respectively. Internal factuality tests also revealed that Sol makes about half as many errors as its predecessor.
Furthermore, OpenAI has incorporated Astra's communication style for clearer, more concise answers with reduced jargon. The release also debuts an improved prompt caching system with higher default hit rates, offering discounts of up to 90% on cached input reads within a 30-minute window, which GitHub reports has significantly accelerated Copilot response times.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment