Anthropic releases Opus 5 with ‘close’ to Fable 5’s capabilities
The Verge Hayden Field ● Covered by 17 sources
Anthropic just dropped Claude Opus 5, a model it says comes close to flagship Fable 5's abilities. It's also half the price and packs tighter safety guardrails, landing right as AI regulation heats up.
Based on reporting by The Verge, Hayden Field — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic didn't pick a quiet week to launch anything. Opus 5 arrived on Thursday, not long after the company's public standoff with the US government and just days after a security incident at OpenAI had the whole industry talking. Against that backdrop, Anthropic says Opus 5 "comes close to the capabilities of Claude Fable 5 in many domains" and is notably better at complex coding work. Fable 5, for context, is the Mythos-class model that drew government scrutiny, got pulled offline for weeks alongside Mythos 5, and came back with beefed-up cyber safeguards.
That Fable 5 episode, and the weeks of back-and-forth between Anthropic and regulators that followed, effectively opened a new chapter of AI oversight under the Trump administration. OpenAI felt the ripple effects too — when it shipped GPT-5.6, the model spent roughly two weeks in a limited preview available only to government-approved entities before wider release. Against that pattern, it's notable that Anthropic didn't describe any comparable holding period for Opus 5. Asked directly whether the administration reviewed the model before launch, spokesperson Danielle Ghiglieri told The Verge only that Anthropic "continue[s] to work with our government partners to conduct their own independent testing," including on Opus 5.
Anthropic is positioning Opus 5 as the workhorse for enterprise customers — strong for knowledge work and biology, and pitched as the model people reach for day to day. Fable 5 keeps its spot as the tool for the biggest, longest-horizon agent projects. But the company leaned hard on the safety framing here, calling Opus 5 its most aligned Opus model yet and the hardest to trick into misuse, with stronger cybersecurity protections than predecessor Opus 4.8.
The pricing details feel deliberate, too. Opus 5 costs the same per million tokens as its predecessor — $5 for input, $25 for output — which Anthropic frames as half of what Fable 5 costs and slightly less than OpenAI's GPT-5.6. That's a pointed move given the recent noise around AI labs quietly tightening usage caps, including at least one lawsuit, and broader worry that rising model costs are getting pushed onto customers. Anthropic is also rolling out a research-preview "Fast mode" for double the price, plus an opt-in fallback that drops users to a lower-tier model automatically whenever Opus 5's safeguards decide to refuse a request.
My take — AI-written commentary, not fact-checked reporting
Calling a new model "close" to your flagship while quietly matching its predecessor's price is a smart way to dodge the AI money squeeze headlines without admitting anything was ever squeezed. The real story here isn't Opus 5's coding chops, it's that Anthropic won't say plainly whether the government got early sign-off, which after the Fable 5 mess is exactly the kind of detail people should be pressing on. Safety branding is cheap; independent testing that customers can actually verify is not.
Read more about this at: The Verge