Anthropic Cuts Prices With Claude Opus 5.5 Release

Anthropic set the list price of Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, down from $5 and $25. Tokens are the small chunks of text AI companies bill by. Re-reading the same material also got much cheaper. But price per token is only half of the bill. Bridgebench, a group that tests AI models, found Opus 5.5 used more than four times as many tokens as OpenAI's GPT-6 Astra to finish the same coding task, so a lower rate does not guarantee a lower total. Anthropic has published its own estimate of what a typical task costs. Anthropic's cost breakdown ↗

Developer Theo Says Opus 5.5 Finished Code Port in 10 Hours

Developer Theo says Claude Opus 5.5 got his large code conversion working in 10 hours, after two OpenAI models stopped partway through. He used the next day to make the code faster. Moving a whole codebase to a new programming language is a long, multi-step job. Theo says generous usage limits are why he gave Claude a job this size. Another developer, 0xSero, says he stopped using the AI models he ran on his own computer and now mostly uses Claude. see Theo's post ↗

Deloitte Tests Find Opus 5.5 Catches More Bugs Than Opus 5

Deloitte's code-review tests, shared by Bubo, found Opus 5.5 caught 72% of planted bugs on its lowest effort setting. Opus 5 caught 56%, even on its highest. The same tests found 44% more concurrency bugs (errors that appear when many tasks run at once) in code Opus 5.5 writes itself. Reviewer Jack Vinijtrongjit says it checks its work by running it. He also says it used several weeks of his usage limits in one day and still ignores some written instructions. Read more ↗

Anthropic Sets Opus 5.5 Default Effort to Medium

Anthropic changed the default effort setting for Claude Opus 5.5 from high to medium. The setting controls how much work the model does before it answers. Users @notomarsol and @pingrishabh say many people who called the model "lazier" were still on medium. They say the gains show up on medium or high, at roughly a quarter of the old cost. Users also report that the old "think step by step" instruction now gives worse results. Read more ↗