Claude Sonnet 5.5 runs over 30% faster, but effort still controls the bill

Anthropic has released Claude Sonnet 5.5 at the same API price as Sonnet 5: $2 per million input tokens and $10 per million output tokens. The company claims more than 30% higher speed and up to 30% lower cost for most work.

The same token rate buys faster completed work

The change does not come from a cheaper token. Anthropic says the model uses fewer tokens for tasks and responds faster, reducing the cost of completed work. Sonnet 5.5 is available through the Claude API as claude-sonnet-5-5 and is also the model offered on the free tier at claude.ai.

A practical test by Simon Willison exposed an important exception. At max thinking effort, the model spent 128,000 tokens, cost $1.28 and failed to produce an SVG. At xhigh effort, it completed the same kind of task in 41 seconds for 5.74 cents.

Teams should price outcomes rather than tokens

For product and engineering teams, end to end workflow throughput is the useful metric. A model that runs 30% faster can reduce waiting in code review, batch processing and agent loops, provided that it also consumes fewer tokens.

Free access gives Anthropic a broader distribution channel for a capable work model. Users can test quality without a paid plan, while companies get a cheaper path from prototype to API deployment.

Maximum effort can consume the promised savings

Willison's test shows why changing the model name in a configuration file is insufficient. A high effort setting can keep searching, burn the efficiency gain and still return no usable result. Vendor benchmarks also cannot predict performance on a particular repository, dataset and tool stack.

Production telemetry will reveal the real price

Teams should compare Sonnet 5 and 5.5 on identical tasks, measuring success rate, latency, tokens per completed job and human intervention. The meaningful signal will be a durable reduction in outcome cost, not one polished benchmark or a successful pelican on a bicycle.

Lilith's verdict

Sonnet 5.5 can push more finished work through the same pipe. Teams that leave effort at maximum without measuring it may watch the meter run while the model is still looking for the pelican's handlebars.