Anthropic on Thursday released Claude Opus 5, a model the company says approaches the capabilities of its most powerful offering — Fable 5 — while carrying a lower price tag, fewer restrictions, and stronger alignment properties than its predecessor. [1]
Near-Flagship Performance at Half the Cost
Anthropic describes Opus 5 as coming "close to the capabilities of Claude Fable 5 in many domains," with particular improvements in complex coding tasks. [1] According to the company's own benchmark testing, Opus 5 actually outperforms Fable 5 on a number of measures — including a test in which the model wrote its own computer vision pipeline in response to an intentionally incomplete prompt. [2]
The company is positioning Opus 5 as the ideal everyday workhorse for enterprise customers, especially for knowledge work and biology-related tasks, while maintaining that Fable 5 remains the better choice for the most ambitious long-horizon AI agent projects. [1] Anthropic priced Opus 5 at $5 per million input tokens and $25 per million output tokens — identical to its predecessor, Opus 4.8, and roughly half the cost of Fable 5. It is also slightly cheaper than OpenAI's GPT-5.6. [1]
The release comes just two months after Opus 4.8 became available on May 28, and follows the June launches of Mythos 5, Fable 5, and Sonnet 5. The only model in Anthropic's lineup still awaiting a fifth-generation upgrade is the lightweight Haiku. [2]
A Lighter Regulatory Footprint Than Fable 5
One of the most consequential distinctions between Opus 5 and its Fable counterpart is the degree to which safety classifiers engage. Anthropic expects those classifiers to trigger approximately 85% less often for Opus 5 than for Fable 5, reflecting the company's view that a less capable model warrants a lighter regulatory touch. [2]
Opus 5 is also exempt from the 30-day data retention policy that applies to both Fable and Mythos — a restriction that had drawn concern from privacy-focused users and developers. [2]
That does not mean Opus 5 is unrestricted. Cybersecurity-related tasks remain subject to meaningful guardrails. For example, the model is blocked from scanning for vulnerabilities in a compiled software binary, though it is permitted to search for vulnerabilities in source code — a distinction Anthropic draws on the grounds that source-code analysis is more likely to serve defensive purposes. [2]
Anthropic also described Opus 5 as "the most aligned Opus model" and said it is "the least susceptible to being tricked into misuse," a claim that carries added weight given the regulatory climate surrounding the company's more powerful models. [1]
Government Scrutiny and a Shifting Regulatory Era
The release lands in an environment shaped heavily by recent friction between Anthropic and federal authorities. Fable 5 — part of Anthropic's Mythos-class lineup — drew government concern after its initial release, leading to the model being taken offline for several weeks alongside Mythos 5 before both were restored with enhanced cyber safeguards. [1] The episode, and the weeks of negotiations that followed, marked what observers have described as a new chapter in AI regulation under the Trump administration.
When asked whether Anthropic had submitted Opus 5 for government review ahead of Thursday's 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." [1] The statement stops short of confirming a formal pre-release clearance process but signals ongoing coordination.
The context also includes a recent OpenAI security incident that dominated industry conversations in the days before Anthropic's announcement, as well as the rollout of GPT-5.6, which underwent a roughly two-week limited preview restricted to government-approved entities before broader release. [1]
New Features: Fast Mode and Automatic Fallbacks
Alongside the model itself, Anthropic introduced two new features aimed at enterprise API customers.
The first is "Fast mode," launching today in research preview, which offers higher response speeds at double the standard price. [1] The second is Automatic Fallbacks — a beta opt-in that routes a request to a less powerful model when a prompt triggers Opus 5's safety classifier, rather than returning an error message. [1][2] The practical effect is that developers who enable the setting will receive a functional response even when their original request crosses a safety threshold, reducing disruption in production environments.
Addressing the AI Cost Squeeze
Anthropic's decision to hold pricing flat at Opus 4.8 levels appears deliberate. The company has faced at least one lawsuit related to allegations of hidden usage caps, and broader industry concern has grown over AI labs passing infrastructure costs onto customers in opaque ways. [1] By keeping per-token pricing unchanged and explicitly marketing Opus 5 as "half the price" of Fable 5, Anthropic is signaling cost predictability to enterprise buyers at a moment when that message carries real commercial weight.
What to Watch
The immediate questions center on how Opus 5 performs in real-world enterprise deployments relative to both Fable 5 and OpenAI's competing offerings, and whether the government's independent testing process results in any further restrictions or public commentary. Longer term, the remaining gap in Anthropic's model lineup — a fifth-generation Haiku — will be worth tracking as the company works to complete its 5-series rollout.
