OpenAI has introduced GPT-6 Sol and GPT-6 Luna, expanding the GPT-6 model family to provide frontier intelligence at different scales and budgets. These models are designed to be more cost-efficient than their predecessors while maintaining high performance in professional work, factuality, coding, computer use, and alignment.
PLUS ULTRAModel ReleasesOpenAIGPT-6 SolGPT-6 LunaGPT-6
OpenAI Introduces GPT-6 Sol and Luna for Improved Cost Efficiency and Intelligence
PLUS ULTRA by Amenoyomi
The new models leverage improvements in caching and inference to reduce costs. OpenAI announced that API prices for Sol and Luna are 50% lower than the previous GPT-5.6 promotional pricing. Additionally, developers can benefit from enhanced prompt caching, which allows for up to 90% discounts on cached input-token reads.
In performance benchmarks, GPT-6 Sol demonstrated significant advancements. On AutomationBench, Sol outperformed Claude Opus 5 in business workflows while costing only 9% of the competitor's price per task. In coding evaluations such as FrontierCode and DeepSWE, Sol showed substantial improvements, matching or exceeding the performance of Claude Fable 5.1 at a much lower cost. GPT-6 Luna also showed significant efficiency gains, matching the capabilities of GPT-5.6 Sol at approximately one-hundredth of its cost.
In terms of factual reliability, OpenAI reports that GPT-6 Sol makes about half as many mistakes as its predecessor on internal evaluations. The models also adopt the improved communication style introduced with GPT-6 Astra, aiming for greater clarity and less jargon in technical and coding conversations.
GPT-6 Sol and Luna are available starting today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access GPT-6 Luna via the desktop app. For developers, the models are available through the OpenAI API as gpt-6-sol and gpt-6-luna.
PLUS ULTRAby Amenoyomi
The introduction of Sol and Luna marks a shift in focus from pushing the absolute boundaries of intelligence to optimizing the cost-intelligence curve. While GPT-6 Astra is positioned as the most intelligent model for the most demanding projects, Sol and Luna are designed to distribute the benefits of that frontier intelligence across different budgets and scales.
The provided benchmarks indicate that the primary value of these models lies in economic efficiency rather than a leap in raw capability over the absolute frontier. For instance, on AutomationBench, GPT-6 Sol outperforms Claude Opus 5 while requiring only 9% of the cost per task. Similarly, on DeepSWE, Sol matches the performance of Claude Fable 5.1 but at approximately 80% lower cost. These results suggest that the models are designed to achieve "equivalent" frontier-class results at a fraction of the previous resource requirement.
This transition to practical scalability is supported by infrastructure improvements in caching and inference. By reducing API prices by 50% compared to GPT-5.6 promotional pricing and offering 90% discounts on cached input-token reads, the focus moves toward making advanced AI sustainable for high-volume, everyday professional workflows where the cost of sustained use is a critical constraint.
Sources
- Introducing GPT-6 Sol and Luna （OpenAI News, 2026-09-22）
- GPT-6 Sol and Luna （Hacker News Frontpage, 2026-09-22）