It also has less restrictive safeguards than the Fable model.
Just as it was starting to feel like Anthropic had begun to neglect its Opus model, the company has released a new version of its one-time flagship LLM. According to Anthropic, Opus 5 offers capabilities that come close to Fable 5, Anthropic’s most powerful commercially available model, at half the price. More on pricing in a bit, but as for performance, Anthropic says Opus 5 is better at knowledge work than any of its past models. It’s also substantially better at complex coding tasks than Opus 4.8. Anthropic claims it’s better for scientific research too, where it excels at tasks like predicting how variations in a protein sequence might affect the way the molecules function.
At the same time, Anthropic says it has managed to make the model more resistant to being tricked. Additionally, the company notes Opus 5 “exhibits the lowest rates of deceptive behavior.” So hooray for that. One important distinction between Opus 5 and Mythos, which is currently only available to a limited number of vetted organizations through Anthropic’s Project Glasswing initiative, is that the company has specifically avoided training the new model on cyber-related tasks. Due to more its powerful capabilities, Opus 5 is broadly better at those tasks than its predecessor, making it more useful for finding cybersecurity vulnerabilities, but the company says Opus 5 is “substantially behind” its flagship model at exploiting those vulnerabilities.
With today’s release, Anthropic expects Opus 5 will become the model most people will reach for when they need to get some work done, and it’s easy to see why the company believes that. One of the frustrations people had with Fable were the safeguards the company imposed. Anthropic implemented a system where Claude would route prompts related to some topics through to Opus 4.8 instead of Fable 5. Many felt the restrictions were heavy-handed, and made Fable feel unusable for some tasks. To that point, Anthropic says “Claude Opus 5’s safeguards are designed to allow beneficial uses of the model in both cybersecurity and biology.” The company has strengthened some of the model’s cyber-related guardrails, but notes it did so along a “narrow range” of specific tasks. “Based on our testing, we expect the classifiers to intervene around 85 percent less often than they do for Fable 5,” Anthropic said. With Opus 5, Anthropic also isn’t including it in its recently announced 30-day data retention policy, which the company introduced alongside Fable and Mythos 5.
As for pricing. Anthropic says API costs for Opus 5 will remain at $5 per one million input tokens and $25 per one million output tokens. Anthropic has also added an “effort” menu for Opus that users can tweak to tell the model whether they want it to be more thorough or fast and efficient to conserve tokens. Had the company announced the new model before China’s Moonshot came out with its Kimi K3 model, which costs $15 per one million output tokens and offers Fable-like performance on some tasks, the reaction to Opus 5 would probably be very positive. Instead, I’m sure people will grumble about pricing. That said, access to Opus 5 is included with all of Anthropic’s paid tiers. Right now, you can’t sign up for one of Kimi’s subscriptions plans, even if you wanted.
Credit: Source link