Anthropic has released Claude Opus 5.5, its new flagship model, cutting list prices to $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Anthropic says typical workloads cost around 40% less than on the previous Opus generation, a bigger saving than the per-token cut alone suggests, pointing to efficiency gains in how the model handles tasks rather than just a price change.
On Terminal-Bench 4.0, a test of agentic coding skill, Opus 5.5 is reported at 66.4%, among the strongest scores of any model compared at launch, though one rival still edges ahead on a related science-focused variant of the same benchmark. Anthropic had external evaluators, including METR and Frontier Design, test the model before release.
The more notable change is security policy. Opus 5.5 is the first Opus model to carry safeguards similar to those Anthropic introduced with Claude Fable 5.1: routine bug-finding and fixing work proceeds normally, but most other cybersecurity tasks are automatically rerouted to the older Opus 4.8 model instead. Verified security professionals can apply for broader access through Anthropic's Cyber Verification Program, which the company plans to expand in the coming weeks. Anthropic also notes that when its safety classifier intervened during testing and handed a task to Opus 4.8, this likely pulled down Opus 5.5's own benchmark scores on cybersecurity-related tests.
Extended thinking, previously optional, can no longer be fully switched off in this release. Anthropic says Sonnet 5.5 and Haiku 5.5 will follow in the coming weeks to complete the family.
Source: MarkTechPost
