Open AI Cuts Prices of Two AI Models by as Much as 80% as Token Price War Intensifies
OpenAI said Thursday that it was sharply cutting the prices of its two newest artificial-intelligence models, GPT-5.6 Terra and GPT-5.6 Luna, only about three weeks after their public release. Enterprise customers are becoming increasingly sensitive to costs. If they cannot see a clear return on investment, they are reluctant to deploy expensive models, putting pressure on OpenAI to meet demand for lower-cost options. The company is also facing competition from Google, Microsoft and other technology giants that have been promoting more cost-effective models. On July 9, OpenAI officially introduced its next-generation GPT-5.6 model family: Sol, Terra and Luna. Sol is the flagship model, with OpenAI highlighting performance improvements in programming, biology and cybersecurity. Terra is designed for everyday work, while Luna is faster and more affordable. OpenAI said Thursday that it would cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%. Luna API pricing is $0.20 per 1 million input tokens and $1.20 per 1 million output tokens. Terra API pricing is $2 per 1 million input tokens and $12 per 1 million output tokens. Pricing for Sol, as well as ChatGPT and Codex subscription prices and quota budgets, will remain unchanged. Using Terra and Luna will consume fewer credits. “Our core strategy remains to improve model capabilities and operating efficiency in parallel, striving to enable each generation of intelligent models to complete more tasks at a lower cost,” OpenAI said in a statement. The token price war is intensifying. The launch of ChatGPT in 2022 sparked an AI boom, prompting U.S. companies to race to deploy the technology and encouraging employees to use it extensively. That gave rise to the so-called “tokenmaxxing” era, when employers encouraged workers to use as much AI as possible without worrying about the cost. But as AI bills continued to surge—with spending at some companies reaching tens of billions of dollars—many businesses began actively controlling expenses. The shift has drawn close attention from AI developers. At the same time, a wave of open-weight models has risen rapidly, with performance increasingly approaching that of leading closed-source models. After Moonshot AI released Kimi K3, OpenAI’s main rival Anthropic introduced its new Claude Opus 5 and described it as the best-performing and most cost-effective product for many applications. Although Claude Opus 5 performed comparably on programming and knowledge-work tasks to Anthropic’s high-end Claude Fable 5, released in June this year, its price was only half as high. Microsoft CEO Satya Nadella also repeatedly emphasized the company’s cost-effective models during the quarterly earnings call on Wednesday. Earlier this week, Microsoft released a cybersecurity model that it described as “cheap but high-performing.” Google also introduced three new models this month designed to beat rivals on cost. The company said its strongest new model, Gemini 3.6 Flash, has a lower per-task cost than Kimi K3.
Enterprise customers are becoming increasingly sensitive to costs. If they cannot see a clear return on investment, they are reluctant to deploy expensive models, putting pressure on OpenAI to meet demand for lower-cost options. The company is also facing competition from Google, Microsoft and other technology giants that have been promoting more cost-effective models.
On July 9, OpenAI officially introduced its next-generation GPT-5.6 model family: Sol, Terra and Luna. Sol is the flagship model, with OpenAI highlighting performance improvements in programming, biology and cybersecurity. Terra is designed for everyday work, while Luna is faster and more affordable.
OpenAI said Thursday that it would cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%. Luna API pricing is $0.20 per 1 million input tokens and $1.20 per 1 million output tokens. Terra API pricing is $2 per 1 million input tokens and $12 per 1 million output tokens.
Pricing for Sol, as well as ChatGPT and Codex subscription prices and quota budgets, will remain unchanged. Using Terra and Luna will consume fewer credits.
“Our core strategy remains to improve model capabilities and operating efficiency in parallel, striving to enable each generation of intelligent models to complete more tasks at a lower cost,” OpenAI said in a statement.
The token price war is intensifying.
The launch of ChatGPT in 2022 sparked an AI boom, prompting U.S. companies to race to deploy the technology and encouraging employees to use it extensively. That gave rise to the so-called “tokenmaxxing” era, when employers encouraged workers to use as much AI as possible without worrying about the cost.
But as AI bills continued to surge—with spending at some companies reaching tens of billions of dollars—many businesses began actively controlling expenses. The shift has drawn close attention from AI developers. At the same time, a wave of open-weight models has risen rapidly, with performance increasingly approaching that of leading closed-source models.
After Moonshot AI released Kimi K3, OpenAI’s main rival Anthropic introduced its new Claude Opus 5 and described it as the best-performing and most cost-effective product for many applications. Although Claude Opus 5 performed comparably on programming and knowledge-work tasks to Anthropic’s high-end Claude Fable 5, released in June this year, its price was only half as high.
Microsoft CEO Satya Nadella also repeatedly emphasized the company’s cost-effective models during the quarterly earnings call on Wednesday. Earlier this week, Microsoft released a cybersecurity model that it described as “cheap but high-performing.”
Google also introduced three new models this month designed to beat rivals on cost. The company said its strongest new model, Gemini 3.6 Flash, has a lower per-task cost than Kimi K3.