Anthropic today introduced Claude Sonnet 5.5. According to the company, the new Sonnet 5.5 "runs 30%+ faster, and costs up to 30% less for most work." The model is priced the same as Sonnet 5, but appears to outperform it across benchmarks and should be cheaper to operate in practice.
Performance and cost claims
- Anthropic states Sonnet 5.5 runs more than 30% faster on average.
- The company also says it can cost up to 30% less for most workloads.
- Although the list price matches Sonnet 5, Sonnet 5.5 reportedly yields better results on many benchmarks.
Token-handling bug and examples
Sonnet 5.5 exhibited the same bug previously seen in Opus 5.5: with an extreme "max" thinking effort setting, the model processed up to 128,000 tokens (which in the cited example cost $1.28) and then ran out of tokens, failing to produce the expected SVG output.
A shorter example from the same reporting shows that with the "xhigh" thinking effort setting the model produced a result in 41 seconds at a cost of $0.0574 (5.74 cents).
On some coding tasks — including several viral 3D animation tricks — Sonnet 5.5 performed nearly as well as Opus 5.5.
Free-tier deployment and a practical test
Crucially, Sonnet 5.5 is now the model powering the free tier on claude.ai. By comparison, OpenAI's ChatGPT free tier uses Luna 5.6, so Anthropic currently offers a free-tier model that the source describes as more capable.
The author ran a prompt on that free tier: "build me an HTML page that renders a three-dimensional pelican riding a bicycle using WebGL". The returned result was a working page — described as a solid effort — demonstrating that the free Sonnet 5.5 instance can generate HTML/WebGL code to render a 3D pelican riding a bicycle.
Upcoming models
Anthropic's announcement reiterates that Haiku 5.5 will be available "in the coming weeks." The author expressed hope that it will be price-competitive with GPT-6 Luna.
Summary
Sonnet 5.5 promises faster and cheaper operation and is now deployed as the default free-tier model on claude.ai. However, a previously observed token-limit bug persists in at least some extreme prompt settings, and Anthropic plans further releases in the near term.
