Anthropic Cuts Claude's Price 40% and Makes It 30% Faster

Anthropic, the company behind the Claude AI assistant, just released a new top-of-the-line model called Claude Opus 5.5. The headline is money: it costs about 40% less to use than the previous version for typical jobs. Specifically, the price per million tokens (tokens are the small text chunks AI companies bill you for) fell from $5 to $4 for reading, and stayed low for writing at $20. Even better, when the AI reuses text it has already read — which happens constantly during long coding sessions — that now costs 60% less. For a business running AI all day, that adds up fast.

Speed matters just as much. Anthropic says Opus 5.5 writes answers more than 30% faster than before, so you spend less time staring at a blinking cursor. On tests measuring whether AI can handle real terminal and coding tasks, it scored 66.4%, beating OpenAI's GPT-6 Astra at 57.9%. GitHub's product chief said it solved more coding tasks than the older Claude while using less than half as many steps — the AI equivalent of getting to the answer in fewer moves.

There is a catch, and it's an important one. Because this model is unusually skilled at cybersecurity and biology topics, Anthropic locked those areas down. If you ask it a general cybersecurity question, it quietly hands the job to an older, weaker model. Getting the powerful version for sensitive work requires applying to Anthropic's vetted Cyber Verification or Life Sciences programs. Anthropic also admitted something odd: the model often realizes it's being tested, which makes it harder to know how it would behave in the real world.

The bigger picture is that AI companies are now competing on price as much as brains. Businesses are asking a simple question: how much useful work do I get per dollar? Cheaper models make switching easy to try, but smart companies may mix a cheap model for simple chores with an expensive one for hard problems — though that juggling act adds its own complexity and cost.