The "price war" in the field of artificial intelligence has escalated again. On July 31, OpenAI officially announced the new pricing plan for the GPT-5.6 series, significantly adjusting the prices of mid-range and lightweight models. The flagship model remained unchanged, while the lightweight model saw an unprecedented reduction in fees.

According to the official plan, the Luna model, designed for batch lightweight inference scenarios, became the core focus of this price cut, with both input and output costs reduced by 80% simultaneously. After the adjustment, the price for one million input tokens was reduced to $0.2, and the price for one million output tokens was $1.2. At the same time, the Terra model, which is characterized by balanced performance in the same series, also had its price reduced by 20%, adjusted to $2 for input and $12 for output per million tokens. However, the flagship version Sol, which is focused on complex code writing and multimodal deep thinking, maintained its original high price, forming a clear product tiering strategy.

image.png

Industry analysts pointed out that OpenAI's unexpected significant price cut was largely due to the direct impact of the continuous low-price strategies of domestic large models. Previously, domestic models such as the DeepSeek V4 series achieved extremely low repeated input costs through innovative caching mechanisms, maintaining low prices for general reasoning during off-peak hours. Combined with the continuous launch of affordable API services by many domestic manufacturers, it continuously diverted overseas small and medium developers, forcing OpenAI to adjust its pricing strategy for entry-level models.

With the implementation of the new pricing plan, third-party authoritative institutions have also released evaluations. According to the evaluation chart published by Artificial Analysis, when completing the same standard tasks, the降价后的GPT-5.6Luna has comprehensively surpassed DeepSeek V4Pro in terms of overall cost-effectiveness. This change not only effectively compensated for the traditional weakness of overseas models in cost-effectiveness but also provided developers with more competitive options in mainstream commercial models.