Anthropic rolls out Sonnet 5.5 update: No price cut, but fewer tokens are required for the same tasks.

Beating AI News Flash: Anthropic has released Claude Sonnet 5.5, continuing to focus on daily work and programming use cases. Its API unit prices remain unchanged from Sonnet 5: $2 per million input tokens and $10 per million output tokens, exactly half the price of Opus 5.5. Anthropic states the new model is over 30% faster in output, typically requires fewer tokens to complete the same task, cutting per-task costs by up to 30%. The most notable improvement is in agent programming. In Terminal-Bench 4.0, which tests models on completing multi-step tasks via the command line, Sonnet 5.5 jumps from Sonnet 5’s 10.3% to 70.6%, surpassing Opus 5.5’s 66.4%. However, among the 8 official benchmarks released, Sonnet 5.5 only leads Opus in this single category; the other 7 still favor Opus 5.5, though most gaps are narrow. Anthropic clarifies that complex, open-ended tasks requiring ongoing judgment remain better suited for Opus 5.5. But actual per-task savings may not be half. Artificial Analysis measured an Intelligence Index of 56 for Sonnet 5.5 at its highest Max reasoning level, versus 58 for Opus 5.5 Max. However, Sonnet 5.5 used an average of ~193,000 output tokens per task, resulting in a per-task cost of $7.60 in this test—higher than Opus 5.5 Max’s $5.98. Whether it saves money depends heavily on the task and reasoning level selected. Claude Sonnet 5.5 is now available on Claude, Claude API, AWS, Google Cloud, and Microsoft Foundry. Haiku 5.5 will launch in the coming weeks.