Anthropic's Opus 5.5 said to cut coding costs well beyond its 20% price drop
The poster says Opus 5.5 cuts coding bills by about 40% through cheaper cache reads and fewer steps, and beats GPT-6 Astra's best score on a code-merge test at roughly a fifth of the cost per task.
Anthropic's new Opus 5.5 is going to cut a lot of AI coding bills by way more than the 20% price drop people are quoting today. That 20% is on regular tokens. When an AI agent works on your code, most of what you're paying for is the agent re-reading the same files again and again, before every single step it takes. That re-reading just got 60% cheaper. The model also needs fewer steps to finish things. One early tester had it fix up a 200,000 line codebase in under 3 hours, and the old Opus took over 20 hours on the same job. Stack those two together and Anthropic says the typical bill drops about 40%. The part I keep coming back to is the OpenAI comparison. There's a test that checks whether an AI's code would actually get approved and merged by a real team. Opus 5.5 on its normal setting beat the best score GPT-6 Astra has ever posted on it, and did it for about a fifth of the cost per task. A better result for 20 cents on the dollar. For most of this AI race, the model that topped the leaderboard was also the one you paid up for. This is the first time in a while the model winning the coding tests is also the cheap one to run, and that's a really hard thing for OpenAI to answer with a price cut of its own.
This post quotes another post: “Anthropic releases Claude Opus 5.5 model”