Anthropic launched Sonnet 5.5 six days after Opus 5.5. The company says it scores 70.6% on Terminal-Bench 4.0, ahead of Opus 5.5’s 66.4% at xhigh effort. It also says Sonnet 5.5 generates output more than 30% faster than Sonnet 5 and uses fewer tokens per task.

Sonnet costs $2 per million input tokens and $10 per million output tokens, half of Opus 5.5’s $4 and $20.

Half the price per token doesn’t guarantee half the bill, though. If a model needs more tokens to finish the job, the savings shrink. Independent testing from Artificial Analysis found exactly that at max effort, with Sonnet 5.5 costing $7.67 per task to Opus 5.5’s $5.98. My testing last week suggests Opus 5.5 was the new budget model. Sonnet 5.5 made me test that again.

I was never a fan of Sonnet. I tried it in the past because it was cheaper than Opus 5, but I always ended up going back to Opus 5 because cheaper isn’t a win when the output isn’t that great. I came in unsure whether Sonnet had a real place, so I tested it against Opus 5.5, the model I’d most likely replace with it.

I called both models through the Anthropic API with identical prompts, adaptive thinking, and the maximum effort setting. I ran each test five times per model to measure consistency and graded every run against a hidden test suite the models never saw. I logged tokens, cost at list price, and time for every run.

Source link