Anthropic Says Opus 5 Almost Matches Its High-Performance Model At Half The Price
It also has less limited protection than the Fable model.
As it begins to feel like Anthropic has begun to neglect its Opus model, the company has released a new version of its one-time flagship LLM. According to Anthropic, the Opus 5 offers capabilities approaching the Fable 5, Anthropic’s most powerful model commercially available, at a fraction of the price. More about price in a bit, but in terms of performance, Anthropic says the Opus 5 is better at information work than any of its previous models. It is also much better for complex coding tasks than Opus 4.8. Anthropic also excels in scientific research, where it excels at tasks such as predicting how variations in protein sequences might affect the way molecules work.
At the same time, Anthropic claims to have been able to make the model resistant to manipulation. Additionally, the company notes the Opus 5 “shows very low rates of deceptive behavior.” So hooray for that. One key difference between the Opus 5 and the Mythos, which is currently only available to a limited number of organizations tested by Anthropic’s Project Glasswing, is that the company specifically avoided training the new model for cyber-related tasks. Because of its powerful capabilities, the Opus 5 is much better at those tasks than its predecessors, making it more useful for detecting cybersecurity vulnerabilities, but the company says the Opus 5 is “significantly behind” its flagship model in exploiting those weaknesses.
With today’s release, Anthropic expects the Opus 5 to be 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. Another frustration people had with Fable was the protections the company put in place. Anthropic implemented a system where Claude moved information related to certain topics to Opus 4.8 instead of Legend 5. Many felt that the limitations were difficult, and made Fable feel unusable for other tasks. At that point, Anthropic says “The Claude Opus 5 defense is designed to allow the beneficial use of the model in both cybersecurity and biology.” The company has beefed up some of the model’s Internet-related monitoring features, but notes that it has done so with a “narrow scope” of specific operations. “Based on our testing, we expect classifiers to fit in about 85 percent less than they do in Fable 5,” Anthropic said. With Opus 5, Anthropic also does not include it in its recently announced 30-day data retention policy, which the company introduced alongside Fable and Mythos 5.
As for the prices. Anthropic says Opus 5 API fees will remain at $5 for 1 million input tokens and $25 for 1 million output tokens. Anthropic also added an “efforts” menu for Opus that users can adjust to tell the model whether they want it to be perfect or faster and more efficient to save tokens. If the company had announced a new model before China’s Moonshot came out with its Kimi K3 model, which costs $15 for a million issuing tokens and offers Fable-like performance in other functions, the reaction to Opus 5 would probably have been very positive. Instead, I’m sure people will complain about the prices. 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 subscription plans, even if you wanted to.
