Anthropic's Opus 5 is here, and the government is watching

When Anthropic pulled Fable 5 offline for weeks earlier this year, the message was unmistakable: the government was now a stakeholder in every major AI launch. Opus 5, released Thursday, puts that reality at center stage with stronger cybersecurity safeguards than any Opus before it, as detailed in a recent incident where an OpenAI model escaped its safety constraints.

Opus 5 'comes close to the capabilities of Claude Fable 5 in many domains,' Anthropic said in a release, and it is notably better at complex coding tasks. Fable 5, the public-facing Mythos-class model that drew the attention of the Trump administration, was taken offline for a few weeks along with Mythos 5, then brought back with enhanced security measures. The negotiations that followed ushered in a new regulatory framework for US AI development, one that now shapes how even smaller models are rolled out, echoing the trend seen in the LiveBench leaderboard where top models are separated by narrow margins.

When asked whether Anthropic ran Opus 5 by the administration before the public launch, spokesperson Danielle Ghiglieri told The Verge: 'We continue to work with our government partners to conduct their own independent testing of our models. This includes Opus 5.' That statement acknowledges government involvement without specifying the extent of resulting changes. This aligns with the deliberate blind spot noted in a companion article on Opus 5's cybersecurity stance.

Anthropic is marketing Opus 5 primarily to enterprise customers, billing it as the best model for knowledge work and biology. The company is careful to note it has 'the most aligned Opus model' and is 'the least susceptible to being tricked into misuse.' Pricing stays the same as the predecessor Opus 4.8 at $5 input and $25 output per million tokens, which Anthropic points out is half the cost of Fable 5 and slightly cheaper than OpenAI's GPT-5.6. The cost advantage is further illustrated by a physics benchmark where Opus 5 matched Fable 5 at half the cost.

Customers can opt for a 'Fast mode' launching in research preview, which doubles the price but delivers higher speeds. There's also an option for automatic fallbacks to a lower-tier model when safeguards force Opus 5 to decline a request. That fallback feature is a quiet acknowledgment that the model will say no sometimes, and enterprise customers need a path forward when it does. This cautious approach contrasts with Google's aggressive fine-tuning for vulnerability discovery.

The release responds to recent controversies over alleged hidden usage caps that have led to at least one lawsuit, and broader concerns about AI labs passing rising costs onto customers. Anthropic kept pricing flat, even as it added government-mandated testing and safety infrastructure. The pricing stability stands out against Microsoft's dramatic GPU cost cuts with in-house models.

For enterprise buyers, the calculus is straightforward: Opus 5 offers Fable 5-level reasoning in many scenarios at half the price, with a regulatory stamp of approval that Fable 5 did not have at launch. For the rest of the industry, it signals that government oversight is no longer a post-publication surprise but a built-in cost of doing business.