OpenAI cut GPT-6 token prices in half. The bigger lever may be the cache.
The New Stack Meredith Shubel ● Covered by 7 sources
OpenAI cut GPT-6 Sol and Luna token prices in half. The bigger savings may come from caching more reused context, not just cheaper tokens.
Based on reporting by The New Stack, Meredith Shubel — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI rolled out GPT-6 Sol and Luna on Tuesday as cheaper versions of GPT-6 Astra. They get closer to Astra on alignment than GPT-5.6 Sol did, but they still don’t reach the flagship model. The headline move, though, was price: the company cut API token rates hard, and it says developers can push costs down further by making better use of caching.
OpenAI says the new models benefit from improvements in caching and inference that let it serve them at lower cost. Compared with GPT-5.6, API prices for Sol and Luna are down 50%, and Luna output tokens are 58% lower. The company also says prompt caching now hits more often by default, which means the model can reuse earlier context instead of reprocessing it from scratch every time.
That matters because the same cached context can shave off both latency and spend. OpenAI says developers can now change reasoning effort and tool availability without breaking the cache, so an agent can work harder on one step, ease up on the next, and still keep earlier context intact. The company also added a Prompt Caching Dashboard so developers can see how much input is being cached, how that changes over time, and where caching opportunities are being missed.
OpenAI says GitHub has already seen the share of prompt tokens needing fresh processing drop by more than 50% across billions of requests over the past several months. That is a pretty blunt answer to the current AI pricing fight: cheaper models help, but so does wasting less work. In the agent era, the bill is not just about what the model reads. It is also about how often it has to read the same thing twice.
My take — AI-written commentary, not fact-checked reporting
This is the right fight. Cutting token prices is the flashy part, but cache discipline is where the adult supervision lives. Anyone selling “cheap AI” without talking about reuse is selling a bike with a tiny fuel tank and a very optimistic brochure.
Read more about this at: The New Stack
