Claude Sonnet 5 narrows gap with premium Opus models

Anthropic has just released Claude Sonnet 5, a new AI model that outperforms its predecessor across all benchmarks—even surpassing the larger Opus 4.8 on a key knowledge work test. The upgrade delivers measurable gains while maintaining a cautious stance on sensitive capabilities.

A step forward in reasoning and efficiency

Claude Sonnet 5 builds on Sonnet 4.6 with improvements that translate into higher scores on standard evaluation tests. More notably, it edges past the company’s own Opus 4.8 on the GDPval-AA v2 benchmark for knowledge work, reaching 1,618 compared to Opus’s lower result. These results suggest that Anthropic is steadily closing the performance gap between its mid-tier and premium model lines without requiring a proportional increase in computational demands.

Balancing capability with responsibility

While the new model demonstrates strong performance, Anthropic emphasizes that its cybersecurity-related scores remain well below thresholds currently restricted by the U.S. government. This positioning appears deliberate, aligning with ongoing public and regulatory scrutiny around AI safety. By keeping potentially sensitive outputs constrained, the company signals awareness of broader concerns without sacrificing measurable progress in general-purpose reasoning and efficiency.

---

Source: The Decoder. AI-assisted editorial synthesis — TechnoExpress.