Anthropic launches Claude Opus 5.5 at $4 per million input tokens

Anthropic is positioning this release around customer complaints, not just raw capability. The company says the Opus 5.5 series primarily addresses feedback on cost, efficiency, and communication quality, especially in financial services, law, and software development, and that earlier Claude models had drawn criticism for formulaic, convoluted writing — sometimes called "Claudish." The pricing move is a response to pressure from OpenAI and especially Chinese AI models, which offer lower performance but cost a fraction as much.

The policy layer is also part of the pitch. Opus 5.5 is the first Opus model with safeguards for cybersecurity, biology, and frontier LLM development on par with Fable 5.1; when those safeguards kick in, Anthropic says requests are routed transparently to another model. Anthropic is also fighting distillation attacks, using thousands of fake accounts to extract capabilities at industrial scale, with a "Preserved Thinking" measure first introduced on Fable 5.1 that applies to API accounts created on or after August 31, 2026, plus watermarking to comply with the EU AI Act.

What the claims ultimately hinge on is whether the cheaper headline price holds up once real workloads run. Opus 5.5 is meant to use fewer tokens and generate output more than 30 percent faster, and Anthropic says subscriber five-hour limits rise 20 percent and stretch 25 percent further overall. Still, whether that translates into the promised cost savings depends on how tasks are set up, and, at least for now, the strongest comparisons on price and performance come from Anthropic's own benchmarks rather than fully independent ones.