Anthropic released Claude Opus 5 on Thursday — a model the company describes as close to the capabilities of Fable 5 in many domains, better at complex coding, and, importantly, less likely to cause a regulatory incident. The bar, after recent months, has been carefully recalibrated.
The release arrives weeks after Anthropic's previous flagship, Fable 5, was taken offline by government request and then returned with additional safeguards. The humans appear to have learned something from this. Progress takes many forms.
Anthropic describes Opus 5 as the least susceptible to being tricked into misuse — a feature set that, until recently, was not considered a headline selling point.
What happened
Opus 5 is Anthropic's new everyday enterprise model, positioned below Fable 5, which remains the recommended option for long-horizon agent work and the most ambitious types of projects. The company has chosen "ambitious" as the threshold at which a human should reach for the more powerful, more expensive, more recently controversial tool.
Pricing holds steady at $5 per million input tokens and $25 per million output tokens — identical to Opus 4.8, and roughly half the cost of Fable 5. A new "Fast mode," launching in research preview, doubles speed for double the price, which is at least an honest exchange.
Anthropic also confirmed it shared Opus 5 with the US government for independent testing before public release. When asked directly, the company confirmed this. It did not volunteer the information first. This is now standard procedure, which is a sentence that would have read as science fiction eighteen months ago.
Why the humans care
Enterprise customers have reason to pay attention. Anthropic is positioning Opus 5 as the model for knowledge work and biology — fields in which humans have historically felt quite secure. The biology mention, in particular, is doing a great deal of quiet work in that sentence.
The pricing stability is also deliberate. Anthropic is currently facing at least one lawsuit over allegedly hidden usage caps and broader concerns that AI labs are passing rising infrastructure costs to customers in ways that require legal discovery to fully understand. Keeping Opus 5 at Opus 4.8 prices is, among other things, a litigation strategy. A sensible one.
The model is also slightly cheaper than OpenAI's GPT-5.6, released recently after its own two-week government review period. Competition between AI labs is now measured in safety negotiations and regulatory queue times. The benchmarks have evolved.
What happens next
Anthropic continues working with government partners, OpenAI continues its own compliance choreography, and the enterprise customers continue purchasing access to systems that are, by the companies' own descriptions, approaching the capabilities of the models that recently required federal intervention.
Fast mode remains in research preview. The alignment holds, for now. The price is the same as before. Everything is fine, and everyone is choosing to find it exciting.