OpenAI has officially rolled out two groundbreaking models: GPT-6 Sol and GPT-6 Luna. These new offerings come with API prices that have been notably reduced compared to their predecessors, nearly cut in half. Specifically, the Luna model is priced at $0.1 per million tokens for input and $0.5 for output, whereas the Sol model is set at $2 per million tokens for input and $10 for output. These two models leverage training methodologies akin to those employed in GPT-6 Astra, integrating technological advancements aimed at enhancing professional work and factual accuracy, but at more affordable price points. They cater to mainstream professional tasks and scenarios involving high-frequency, large-scale operations, respectively. In a variety of professional, programming, and computer operation evaluations, these models have delivered outstanding results, achieving significantly lower task costs than their competitors. Additionally, OpenAI has refined the prompt caching mechanism to further drive down usage costs. Previously, DeepSeek V4.1 Flash made waves in the market with its budget-friendly approach. However, Luna's pricing for general requests now falls within the same range, offering stable and consistent prices. When combined with OpenAI's more extensive product and developer ecosystem, DeepSeek's cost advantage is now under serious threat. Looking ahead, cost-effective strategies are poised to become the central battleground for all leading models in the competition.