OpenAI Goes After AI Costs With GPT-6 Sol and Luna, Slashing Prices by 50%

OpenAI is cutting the cost of running its latest AI models as the company and rival Anthropic increasingly compete to make sophisticated AI agents cheaper to deploy at scale.

OpenAI on Tuesday introduced GPT-6 Sol and GPT-6 Luna, two lower-cost additions to its GPT-6 lineup, and said it would cut API prices for the models by up to 50% compared with GPT-5.6’s promotional pricing.

The move comes hours after Anthropic unveiled Claude Opus 5.5, the first model in its new Claude 5.5 family. Anthropic said Opus 5.5 costs 40% less to run than its predecessor while improving performance on coding, computer use, and knowledge-work tasks. 

The announcements underscore a shift in the AI model race toward cost-per-task and efficiency, particularly as companies deploy AI agents that run for extended periods and consume large numbers of tokens.

OpenAI said GPT-6 Sol will cost $2 per million input tokens and $10 per million output tokens, down from $4 and $20, respectively. GPT-6 Luna will cost 10 cents per million input tokens and 50 cents per million output tokens, compared with 20 cents and $1.20 previously.

OpenAI is positioning GPT-6 Astra as its highest-performance model, while Sol and Luna are aimed at workloads where speed and operating costs matter more than squeezing out the highest possible benchmark score.

The company said GPT-6 Sol scored 33.2% on AutomationBench at its highest reasoning setting, at a reported cost of 27 cents per task. OpenAI compared that with GPT-6 Astra, which scored 30.3% at low effort but cost 3.9 times as much per task.

The companies are also emphasizing how their models perform on longer-running agentic tasks rather than simply traditional question-and-answer benchmarks.

OpenAI said GPT-6 Sol made about half as many errors as its predecessor in an internal factuality test based on de-identified conversations in which users had flagged mistakes. It also introduced expanded prompt-caching tools aimed at reducing the cost of applications that repeatedly send the same context to a model.

Anthropic is taking a similar approach. The company said cache reads account for a majority of the costs of agentic and coding workloads and cut its cache-read price by 60% from Opus 5.

The back-to-back launches suggest the AI model race is increasingly becoming a contest over economics as well as capability. As companies move from experimenting with chatbots to running AI agents across coding, research and other workflows, the cost of every task — and not just the quality of the final answer — is becoming a bigger part of the equation.

For OpenAI and Anthropic, that puts pressure on each new generation of models to deliver more capable systems without making them prohibitively expensive to run at scale.

Photo: Shutterstock