When AI Gets Cheaper by Getting Smarter

Most AI price drops come from price wars — one lab cuts rates, others follow. Anthropic's Sonnet 5.5, released yesterday, is different. The per-token price stayed exactly the same. Instead, the savings come from efficiency: the model uses fewer tokens — small chunks of text an AI charges you for — to finish the same job.

Here is a real example. A task that consumed 400,000 input tokens on Sonnet 5 shrank to 280,000 on Sonnet 5.5. The price per token did not change — $2 per million input, $10 per million output — but the total bill fell from $1.40 to $0.98 on that one task. At hundreds of tasks a day, that compounds.

The jump in coding ability is the most striking number. Sonnet 5.5 scored 70.6% on Terminal-Bench, a test for AI agents doing real coding work — compared to just 10.3% for the previous Sonnet. Fewer wrong moves means fewer retries, and fewer retries means fewer tokens burned.

This makes Sonnet 5.5 ideal for everyday workhorse tasks: fixing bugs, reviewing code, writing documents, and handling support tickets. Zendesk reported 20% faster ticket resolution in early tests. The model is available now on Amazon Web Services, Google Cloud, and Microsoft Azure.

If you are running AI workflows at any volume, test Sonnet 5.5 on your actual tasks today. The gains stack: faster output, fewer tokens, lower bills — all at the same price per token.