Do Son 
Many observers assumed Anthropic had quietly sidelined the Opus line. That assumption now looks premature. The once-flagship series has returned with a major upgrade.
Anthropic says the new Claude Opus 5 nearly matches Fable 5, its strongest commercial model on the market today. Yet Opus 5 costs roughly half as much to run.
Sharp Gains in Knowledge Work and Coding
On performance, Anthropic claims Opus 5 surpasses every earlier model at knowledge work. Complex coding tasks also improve markedly over the previous Opus 4.8. Furthermore, the model handles scientific research more capably.
For example, it predicts more precisely how mutations in a protein sequence will affect molecular behaviour.
A Stronger Safety Profile
Security also gets attention. Anthropic stresses that Opus 5 resists malicious deception more effectively. In fact, the company reports its lowest observed rate of deceptive behaviour.
Anthropic also drew a deliberate line. Unlike Mythos, the restricted model available only to a few vetted organisations, Opus 5 avoids specialised training for cybersecurity tasks.
Its strong foundations still help it find vulnerabilities better than earlier models. However, when it comes to actually exploiting those flaws, Opus 5 trails far behind the flagship Fable line.
Loosening the Fable 5 Safety Shackles
Anthropic expects Opus 5 to become most people’s default model for daily work. One reason stands out. The company has eased the safety limits that users often criticised in Fable 5.
Previously, many complained that Fable 5 guarded too aggressively. The system frequently downgraded certain prompts and routed them to Opus 4.8 instead. As a result, Fable 5 felt awkward on some tasks.
Anthropic has now redesigned the safety mechanism in Opus 5. The new design permits beneficial use across cybersecurity and biology.
Consequently, the company estimates that safety-classifier interventions fire about 85% less often than on Fable 5. Opus 5 also escapes the 30-day data-retention policy that Anthropic applied to Fable and Mythos 5.
Pricing Holds Steady and Adds an Effort Dial
On cost, Opus 5 keeps the same rate card as Opus 4.8. API calls run $5 per million input tokens and $25 per million output tokens. By comparison, Fable 5 charges $10 for input and $50 for output.
Access to Opus 5 already comes bundled with every paid Anthropic subscription. Moreover, Anthropic has added a new Effort menu.
The setting lets you steer the model directly. You can ask it to think more thoroughly, or you can favour fast, efficient output to conserve tokens.
A Sweet Spot Under Pricing Pressure
Relaxed limits and an 85% drop in safety interventions rank among the model’s biggest draws. Even so, Anthropic still faces real market pressure on price.
Opus 5 defends its half-of-Fable-5 position. Yet a rival complicates the picture. Moonshot AI, the Chinese startup behind the buzzy Kimi K3, charges just $15 per million output tokens and reportedly rivals Fable on some tasks.
Against that benchmark, Opus 5 still looks a little expensive. However, Kimi K3 has not yet opened broad paid subscriptions. Therefore, as a ready-to-use, world-class model, Opus 5 remains the best current balance of performance, cost, and practicality.
Support Our Threat Intelligence
If you find our CVE report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
