Anthropic unveils Opus 5
What's the story
Anthropic has unveiled its latest artificial intelligence (AI) model, Opus 5. The new addition to the company's flagship series is smaller than Fable 5 and comes with a lower price tag and fewer restrictions. This makes it a more suitable option for most applications. Interestingly, Opus 5 outperforms Fable 5 in several benchmarks mentioned in the official announcement.
Enhanced performance
Significant improvements over its predecessor
Opus 5, which was launched just two months after its predecessor Opus 4.8, is said to be "much stronger at verifying its work and iterating carefully until it succeeds."
This claim is backed by benchmark testing where Opus 5 successfully wrote its own computer vision pipeline in response to an incomplete prompt.
Unlike Fable and Mythos, Opus 5 isn't burdened by a 30-day data retention policy that had raised privacy concerns among some users.
Safety measures
New beta feature called Automatic Fallbacks
Anthropic expects the classifiers to engage 85% less often for Opus 5 than they will for Fable 5. This is due to a lighter touch given to the model.
To make safeguards less disruptive when they do engage, Anthropic has launched a beta feature called Automatic Fallbacks.
It automatically routes requests to a less powerful model when a prompt triggers the safety classifier.
Strategic launch
Pricing of Opus 5
Anthropic is primarily targeting its enterprise customers with the new model, marketing it as the best fit for knowledge work and biology.
The company claims Opus 5 is "the most aligned Opus model and is the least susceptible to being tricked into misuse."
In terms of pricing, per million tokens cost for Opus 5 remains unchanged from its predecessor ($5 input/$25 output) but is half that of Fable 5.