OpenAI 80% price reduction in one go, GPT-5

The price of OpenAI has been reduced by 80% in one go, and GPT-5.6 Luna is more cost-effective than DeepSeek V4 Pro. OpenAI officially announced a new pricing plan for the GPT-5.6 series. This price adjustment is only for the two mid-range and lightweight models of Terra and Luna. The flagship GPT-5.6 Sol price remains unchanged. Among them, Luna, which is designed for batch lightweight inference scenarios, has experienced the largest price reduction in history, with input and output costs reduced by 80% simultaneously. Before the price adjustment, Luna’s input Token per million was US$1 and output was US$6. After the adjustment, the input was only US$0.2 (approximately RMB 1.4) and the output was US$1.2 (approximately RMB 8.1). The balanced version of Terra in the same series was simultaneously reduced by 20%, with an input price of 2 US dollars and an output price of 12 US dollars, while Sol, which focuses on complex code and multi-modal in-depth thinking, maintained its original price, forming a clear layered strategy of "low-price Luna grabs the market, and high-end Sol maintains profits". Comprehensive performance-to-price ratio, making up for cost shortcomings. OpenAI simultaneously released evaluation charts from the third-party organization Artificial Analysis, comparing mainstream lightweight commercial models from the two dimensions of unit task cost and model comprehensive score. The data shows that under the new pricing, Luna's comprehensive price-performance ratio for completing tasks of the same standard exceeds that of DeepSeek V4 Pro. It scores higher and costs less in scenarios such as general text creation, multi-round conversations, and tool invocation. Previously, the biggest shortcoming of overseas models compared with domestic cost-effective products was cost, and this price reduction directly makes up for it. Industry analysts believe that the fundamental driving force for the sharp price reduction comes from the continued low-price impact of domestic large models - DeepSeek V4 Pro relies on the caching mechanism to achieve extremely low repeated input costs. Conventional inference pricing during off-peak periods remains low for a long time. Multiple domestic manufacturers continue to launch affordable API services, which continuously diverts overseas small and medium-sized developer customers. When the low-price advantage of domestic models is no longer a moat, the next war in the large model industry may have just begun. via AI News (author: AI Base)