Anthropic's Claude Sonnet 5.5: 30%+ faster, nearing Opus 5.5

Anthropic's release of Sonnet 5.5 follows its previous Sonnet 5, and the new model is positioned as both faster and cheaper on a per-task basis. While the per-token list price remains the same as Sonnet 5, the company and early testers indicate that fewer tokens are needed per task, lowering overall cost. For example, Balyasny Asset Management found token use per response dropped to about one-quarter of Sonnet 5's level on financial tasks.

The model's benchmark scores show significant gains over Sonnet 5 and come close to Opus 5.5 on many tests, even exceeding it in command-line programming. This suggests Anthropic is aiming to blur the line between its mid-tier and high-end models for routine work. Early users like Base44, Zendesk, and Box report practical improvements in speed and token efficiency, though long-chain reasoning remains a clear Opus 5.5 advantage.

The move matters for developers choosing models for coding and API use, as the combination of lower cost per task and near-Opus performance could change default choices. The outcome hinges on whether teams prioritize speed and cost over depth for their specific workloads, and how Sonnet 5.5's safety restrictions—such as switching to Sonnet 5 for high-risk cybersecurity requests—affect real-world use.