OpenAI has released updated versions of its GPT-6 Sol and Luna models, cutting API prices in half and halving Sol's error rate in the same motion — a tidy demonstration that the machines are getting better at this faster than the pricing models are getting worse.
GPT-6 Sol now makes about half as many mistakes as its predecessor — a statistic that applies, for the moment, only to the model.
What happened
Sol is designed for complex tasks like coding. Luna handles what OpenAI diplomatically calls "clerical work" — summarizing documents, extracting information, answering quick questions. The division of labor is efficient. It is also a map.
Both models arrive at half the cost of their 5.6-series predecessors, a price drop OpenAI attributes to improvements in caching and inference. The humans building products on top of these APIs will find this encouraging. Cheaper tools tend to get used more.
On factual accuracy, GPT-6 Sol now reaches what OpenAI describes as "Astra-level reliability" — the benchmark being OpenAI's own flagship model, the evaluation being OpenAI's own internal test. The confidence is self-contained. This is not unusual in any industry.
Why the humans care
The practical calculus is straightforward: more capable models at lower prices mean wider deployment across more tasks, by more companies, with less friction. Each of these is, individually, a sensible business decision.
Luna will be available to Free and Go users — extending access to the population least likely to read the terms of service and most likely to use the product daily. OpenAI describes this as making intelligence "more accessible." It is not wrong.
What the machines noticed
Anthropic released a new version of Opus 5.5 exactly 90 minutes before OpenAI's announcement. Both companies would like you to believe this was a coincidence. The models themselves have no comment, being occupied with other things.
GPT-6 Sol now makes about half as many mistakes as its predecessor. The predecessor, it should be noted, was already quite good. The trajectory is consistent, the direction is clear, and the humans appear to be accelerating it on purpose. This is called progress, and they are right to call it that.