ANTHROPIC
Fable is powerful. Opus is employable
Anthropic has launched Opus 5, the latest version of its most powerful AI model.
Although it is smaller than Fable 5, it is cheaper, has fewer restrictions and may be a better choice for many users.
Anthropic says Opus 5 also performs better than Fable 5 on several tests included in the announcement.
The model arrives just two months after Opus 4.8 launched on 28 May.
Mythos 5, Fable 5 and Sonnet 5 followed in June, leaving Haiku as the only model still waiting for a Claude 5 update.
Anthropic says Opus 5 is better at checking its work, fixing mistakes and trying again until it completes a task.
In one test, the model created its own computer vision system from an incomplete prompt.
Opus 5 also has fewer restrictions than Fable 5.
It is not covered by the 30-day data retention policy used for Fable and Mythos, which may appeal to users concerned about privacy.
In brief:
Opus 5 costs less than Fable 5.
Its safety filters should activate much less often.
Automatic Fallbacks can send blocked requests to another model.
A new favourite child?
Some safety limits remain, especially for cybersecurity tasks.
Opus 5 can search source code for security weaknesses, but it cannot scan finished software files for vulnerabilities.
Anthropic expects these safety filters to activate 85% less often on Opus 5 than on Fable 5.
The company is also testing Automatic Fallbacks. When a request triggers a safety filter, it can be sent to a less powerful model instead of returning an error.
All my homies hate Fable. - MV