Anthropic cuts Claude Opus 5.5 price 20%, OpenAI answers with cheaper Sol and Luna

The two releases landed minutes apart, which is itself the story: Anthropic and OpenAI are competing on price as much as on capability. Anthropic framed Opus 5.5 around efficiency, noting that the $4 input and $20 output pricing works out to 40% less on a typical workload than Opus 5, with cache reads falling 60% to 20 cents. OpenAI aimed lower down the cost curve entirely, halving the price of both new models against the GPT-5.6 versions that carried the same names and discounting cached input reads by 90%. Both companies also emphasized brevity, with Anthropic saying the model puts the most important information up front and OpenAI saying its answers come back shorter with less jargon.

The published benchmark figures point in different directions, and neither launch offers a clean comparison against the other. Anthropic reported wide gaps over Opus 5 on scientific research, where Terminal-Bench-Science 0.1 gave Opus 5 29% and the new model 58.7%, and on AutomationBench, where task completion rose to 40% from 26.9%. OpenAI reported Sol at 68.8% on the DeepSWE v1.1 software engineering test, which it said is within 1.1 points of the earlier Claude Fable 5, and Luna at 66.6% on the same test. Because OpenAI's comparisons run against Fable 5.1 and Opus 5, buyers cannot read a head-to-head out of either announcement.

Anthropic also kept restrictions in place, routing most cybersecurity tasks to the older Opus 4.8 model and fencing off high-risk biology work behind verification programs that now include a life sciences track. For the business reader, the practical question is whether the lower prices hold and how quickly Anthropic's promised Sonnet 5.5 and Haiku 5.5 follow, since the cost of running routine coding and agent workloads is the line item these launches most directly touch.