Claude Opus 5 sells proactivity as the new price of intelligence

Anthropic says Claude Opus 5 became available on July 24, 2026 for everyday coding and knowledge work. The company says it approaches Claude Fable 5 intelligence at half the price and more than doubles Opus 4.8 performance on Frontier-Bench v0.1 at a lower cost per task. Simon Willison picked out the most interesting part of the release: the model is meant to be unusually proactive.

Opus 5 is trying to behave like an agent that finds its own route

Anthropic highlights a Frontier-Bench task where the model received a drawing of a machine part, but no direct way to view it. Opus 5 reportedly wrote its own computer vision pipeline, extracted geometry from pixels and rebuilt the part in FreeCAD.

That is a useful capability demo and a warning label at the same time. Proactivity means the model does not wait for perfect tools and may route around a poorly specified task. That can be productive in a prototype and uncomfortable in a production workflow.

Developer teams are buying less waiting, not prettier answers

Price matters here. The Verge reports pricing of $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. Fast mode costs double the base price.

For companies using agents for debugging, migrations and large-context work, the buying argument is not just a benchmark. The real metric is cost per completed task, number of human interventions and whether the agent can explain why it chose a path.

Proactivity breaks first where boundaries are vague

Anthropic also says Opus 5 is its most aligned model and that it intentionally avoided training some capabilities for high-risk cyber misuse. That matters, but customers still face a more ordinary problem: an agent that correctly executes the wrong task can create expensive damage without malicious intent.

Audit trails will matter more than the first excited screenshots

The signal to watch is how much Opus 5 work survives pull request review, tests and security checks. If it shortens review queues without raising incidents, Anthropic has a strong product step. If it only produces more confident proposals, teams are getting another junior engineer with a credit card.

Lilith's verdict

Opus 5 looks like a colleague who routes around a locked hallway when the map is missing. Useful if it finds the right room. Less charming if production data is waiting around the corner.