Just two months after Opus 4.8, Opus 5 arrives: it beats Fable 5 on several benchmarks, costs less and operates with fewer restrictions — its safety classifiers trigger 85% less often. It also introduces "Automatic Fallbacks", routing requests to smaller models instead of returning an error.
Anthropic introduced Opus 5 on July 24 as the new version of its flagship line, described as "much stronger at verifying its work and iterating carefully until it gets there". In the demos it wrote computer-vision pipelines from incomplete prompts and searched source code for vulnerabilities for defensive purposes.
More capable, less restricted: it beats Fable 5 on several benchmarks while costing less, and its safety classifiers fire 85% less often than Fable 5's. It is also not subject to the 30-day data retention policy that covers the Fable and Mythos models.
Automatic Fallbacks (beta): when a safety trigger fires, the request is automatically routed to a less powerful model instead of returning an error — a working answer rather than a red screen.
The context: with this launch, Haiku is the only model left to make its jump to the Series 5.
Go to TechCrunch's coverage →Source: TechCrunch
Go to the original source →