For a while, it looked like Anthropic had quietly moved on from its Opus line. The company’s attention seemed to shift toward Fable and the highly restricted Mythos model, leaving Opus fans wondering what came next. Now Anthropic has answered that question with Opus 5, a model it says can nearly match its most powerful commercially available model, Fable 5, at half the cost.
According to Engadget, Anthropic is positioning Opus 5 as the go-to model for most professional use cases. It performs better than any previous Opus model on knowledge work, handles complex coding tasks significantly better than Opus 4.8, and shows particular strength in scientific research. One specific example the company highlights is predicting how changes in a protein sequence might affect how molecules function, which is the kind of specialized task that matters a lot in biotech and drug discovery.
Anthropic also says the model is harder to trick and shows the lowest rates of deceptive behavior of any model it has released. That’s a meaningful claim in a market where users and regulators are paying close attention to AI reliability and honesty.
One of the bigger stories here is what Opus 5 means for users who found Fable frustrating to work with. Fable came with a routing system that automatically redirected certain prompts to Opus 4.8 rather than letting Fable handle them directly. Many users felt those restrictions were too heavy-handed and made the model feel limited for real-world tasks. Anthropic is addressing that directly with Opus 5, saying its safeguards are designed to “allow beneficial uses of the model in both cybersecurity and biology.” The company says its classifiers will step in about 85 percent less often than they do with Fable 5, which is a significant shift in how the model is expected to behave in practice.
The cybersecurity angle is worth paying attention to. Anthropic deliberately avoided training Opus 5 on cyber-specific tasks, which is the same approach it took with Mythos. The result is that while Opus 5 is better than Opus 4.8 at finding vulnerabilities because of its raw capability improvements, it is “substantially behind” Fable 5 when it comes to exploiting those vulnerabilities. That’s a deliberate design choice, not a limitation Anthropic is trying to hide.
Mythos, for comparison, is currently only available to a small group of vetted organizations through Anthropic’s Project Glasswing initiative. Opus 5 is much more broadly accessible, which makes the cybersecurity guardrail decision relevant for a wider audience.
On the pricing side, Anthropic is keeping API costs at $5 per million input tokens and $25 per million output tokens. The company is also adding an “effort” control that lets users tell the model whether to prioritize thoroughness or speed and token efficiency, depending on the task.
That pricing might have landed better a few weeks ago. China’s Moonshot recently released Kimi K3, which costs $15 per million output tokens and reportedly matches Fable 5 on some benchmarks. That puts Opus 5’s $25 per million output token price in a tougher spot competitively. There will be grumbling. That said, Opus 5 access is included in all of Anthropic’s paid subscription tiers, and Kimi doesn’t currently offer a consumer subscription plan at all, which limits a direct apples-to-apples comparison for most users.
One other detail worth noting: unlike Fable and Mythos 5, Opus 5 is not included in Anthropic’s recently announced 30-day data retention policy. That distinction may matter to enterprise customers who have specific data handling requirements.
The broader picture here is that the AI model market is getting increasingly competitive on price and capability at the same time. Companies like Anthropic are being pushed to offer more performance at lower cost, while also managing growing concerns about safety and misuse. Opus 5 is Anthropic’s attempt to thread that needle for the majority of its users, and whether the pricing holds up against international competition is a question the next few months will answer.
