Anthropic updates Sonnet 5.5: Price unchanged, but it uses fewer tokens for the same work

Beating AI News Flash: Anthropic releases Claude Sonnet 5.5, continuing its focus on everyday work and coding. API pricing remains unchanged from Sonnet 5, at $2 per million tokens for input and $10 per million tokens for output — exactly half that of Opus 5.5. Anthropic says the new model outputs more than 30% faster, typically requires fewer tokens to complete the same task, and can cut per-task costs by up to 30%.

Agentic coding showed the most dramatic improvement. In Terminal-Bench 4.0 (which has models complete multi-step tasks in the command line), Sonnet 5.5 jumped from Sonnet 5's 10.3% to 70.6%, also surpassing Opus 5.5's 66.4%. However, among the 8 benchmarks officially published, Sonnet 5.5 leads Opus in only this one category, with Opus 5.5 still ahead in the other 7 — though most gaps are now quite small. Anthropic also explicitly states that complex, open-ended tasks requiring sustained judgment remain better suited to Opus 5.5.

But on a per-task basis, the cost savings may not amount to half. Artificial Analysis measured at the highest Max thinking tier that Sonnet 5.5 scored 56 on the Intelligence Index, while Opus 5.5 Max scored 58. However, Sonnet 5.5 used an average of about 193,000 output tokens per task, bringing its per-task cost in this test suite to $7.60 — actually higher than Opus 5.5 Max's $5.98. Whether it actually saves money depends heavily on the task and thinking tier.

Sonnet 5.5 is now available on Claude, Claude API, AWS, Google Cloud, and Microsoft Foundry. Haiku 5.5 will be released in the coming weeks.