On 30 July, OpenAIreduced the prices of two models from the GPT-5.6 family, just three weeks after their launch. The cost of using the cheapest model, Luna, fell by 80 per cent, to $0.20 per million input tokens and $1.20 per million output tokens. Terra became 20 per cent cheaper, falling to $2 and $12 respectively. The price of the flagship Sol model remained unchanged. The company attributes the decision to improvements in the efficiency of its infrastructure and models.
This move shows that in 2026, competition in the AI market is increasingly centred not only on quality, but also on the cost of performing a specific task. Companies implementing generative AI are measuring their return on investment more closely, reducing token consumption and combining models from different providers. Rising costs are particularly evident with AI agents that perform multiple operations within a single command. According tothe *Financial Times*, some organisations are already introducing budget caps and scaling back agent projects due to costs.
The price cut boosts OpenAI’s appeal for mass-market applications such as customer service, document analysis and code generation. However, it does not automatically guarantee a price advantage. Anthropic is currently offering the Claude Sonnet 5 model at a promotional price of $2 per million input tokens and $10 per million output tokens until 31 August 2026.
The most significant consequence may be further pressure on providers’ margins and the faster adoption of smaller models. For customers, this means greater choice, but also the need to monitor usage more closely. The price per token is falling, but the total bill may still rise as the scale of AI usage increases.
