OpenAI and Anthropic Cut AI Prices, Easing the Cost of Repetitive Work

OpenAI and Anthropic cut the price of their AI models in a series of moves between July and September 2026. OpenAI dropped the price of GPT-5.6 Luna by 80% on July 30. Anthropic lowered the fee for re-reading previously submitted input with Claude Fable 5.1, released on September 1. With Claude Opus 5.5, unveiled on September 22, it also cut input and output rates. All of these are prices for the API, the channel developers use to plug these models into their own services.

Which models got cheaper, and by how much

OpenAI’s 80% cut applies to GPT-5.6 Luna. After the reduction, the price is $0.20 per million input tokens and $1.20 per million output tokens. A token is a unit of text that an AI model processes. Input refers to what the model reads; output is the answer it produces. In the same announcement, the price of Terra fell 20%.

September brought lower prices for successor models as well. GPT-6 Sol, released on September 22, is priced at $2 per million input tokens and $10 per million output tokens. That is 50% cheaper on both input and output than the discounted price of its immediate predecessor, GPT-5.6 Sol. GPT-6 Luna comes in at $0.10 for input and $0.50 for output on the same per-million basis. GPT-6.1 Sol, released September 29, kept Sol’s input and output prices unchanged.

With Claude Fable 5.1, Anthropic sharply reduced the fee for re-reading cached input. Announced on September 1, that rate is $0.25 per million tokens, 75% cheaper than Fable 5. Anthropic estimated that the total bill users actually pay would fall by roughly 25% on typical tasks and by as much as roughly 45% on complex work handled by agents, AI systems that carry out tasks on their own.

Why the cost of reading the same thing again went down

Both companies said they cut prices because the cost of running the models has come down. OpenAI said it had improved the efficiency of the system that generates responses and of the way stored content is reused. Anthropic said Claude Opus 5.5 runs with less computation than Opus 5, and that the cost of re-reading earlier input is lower as well.

Opus 5.5 continued the downward trend. On September 22, Anthropic cut the model’s input and output prices by 20% each compared with Opus 5. The fee for re-reading earlier input came down 60%. Anthropic estimated total cost on typical tasks would drop by about 40%. GPT-6.1 Sol also lowered the price of re-reading stored input to $0.10 per million tokens.

These changes are relevant to work that involves reading earlier conversations or the same document over and over. The charge applied when already-processed content is reused has gotten cheaper. For services that maintain context across long conversations or refer to the same material repeatedly, this fee shows up on the bill alongside the standard input rate.

Where this is available in South Korea

South Korea is a supported region for both the OpenAI API and Anthropic. OpenAI made GPT-6 Sol and Luna available through the API as well as ChatGPT Work and Codex. At launch they were not available in the standard Chat interface, and the same applied to GPT-6.1 Sol. Anthropic said Opus 5.5 can be used on the Claude Platform and through Amazon Web Services (AWS), Google Cloud and Microsoft Azure.

Anthropic says Claude supports a Korean-language interface and conversations in Korean. OpenAI likewise documents its models’ multilingual capabilities. The cuts affect costs for users who connect AI to their own services or workflows. How much anyone saves depends on which model they use, which products they have access to, how much input and output they generate, and how much of their usage involves re-reading earlier content.