Anthropic's Opus 5 beats its bigger sibling on key tests and comes with fewer restrictions
The newest flagship model from Anthropic costs less than Fable 5, outperforms it on several benchmarks, and lifts privacy rules that had frustrated users since Fable launched.
Key points
- Anthropic launched Opus 5 on Friday, two months after Opus 4.8 became available on 28 May.
- Opus 5 outperforms Fable 5 on multiple benchmarks, the company's own tests show.
- Unlike Fable 5 and Mythos 5, Opus 5 is not subject to Anthropic's 30-day data retention policy.
- Anthropic says its safety filters will trigger 85% less often on Opus 5 than on Fable 5.
- A new opt-in feature called Automatic Fallbacks gives developers a usable answer instead of a blank error when a prompt is blocked.
Anthropuc's newest AI model, Opus 5, arrived on Friday, and its headline number is a surprise: it outperforms Fable 5, the company's heavier, more expensive model, on several of its own benchmark tests.
Benchmarks are standardised problem sets, think timed exams for AI, that researchers use to compare models. Scoring higher on them does not guarantee better performance in every situation, but they are the clearest yardstick the industry has.
So why does the cheaper model beat the expensive one?
Anthropus built Opus 5 to check its own work more carefully and keep trying until it gets an answer right. In one test cited in the launch announcement, the model wrote its own computer-vision pipeline, software that lets a machine interpret images, in response to a deliberately vague prompt, then revised it without being asked.
Fable 5 was positioned as Anthropic's most capable model, but Opus 5 appears to have closed or crossed that gap in specific areas while costing less to run.
What about privacy and restrictions?
Opus 5 skips the 30-day data retention rule that applies to Fable 5 and Mythos 5. Data retention means the company holds copies of your conversations for a set period. Removing that rule matters most to businesses and professionals who handle sensitive information.
There are still guardrails in place, particularly around cybersecurity. Opus 5 will not scan a compiled software binary, a finished program, for security holes, because that process most often serves attackers. It will, however, search source code, the human-readable instructions a programmer writes, for the same vulnerabilities, because that task is more commonly used to fix problems than to cause them.
Broadly, Anthropic expects Opus 5's safety filters to activate 85% less often than those on Fable 5.
What is the new Automatic Fallbacks feature?
Automatic Fallbacks is a practical fix for developers, people who build apps and tools on top of Anthropic's models. When a prompt trips a safety filter, the system used to return an error message, a dead end. With Automatic Fallbacks switched on, the request is quietly rerouted to a less powerful model that can still answer, so the app keeps working.
Users of consumer products are unlikely to notice it directly. Developers building on Anthropic's API, the technical interface that lets software talk to the model, can opt in now through a beta programme.
Where does this leave Anthropic's model range?
Opus 5's launch means the 5-series now covers Mythos 5, Fable 5, Sonnet 5 and Opus 5. The only model still waiting for a 5-series upgrade is Haiku, the lightweight option aimed at simpler, faster tasks.
First reported by TechCrunch AI, the launch leaves Anthropic's lineup looking more competitive at every price point, with the heaviest workloads no longer automatically routed to the most expensive option.
