“Sonnet is really for the cost-conscious customer where they might not need as much intelligence,” Theo Chu, a research product manager at Anthropic, told CNBC. “It might be routine tasks that just need execution, but don’t need that judgment that Opus can bring.”
Anthropic says Sonnet 5.5 runs more than 30% faster than Sonnet 5 and can cost up to 30% less per task because it typically uses fewer tokens. Pricing remains at $2 per million input tokens and $10 per million output tokens, compared with $4 and $20 for Opus 5.5.
The company is highlighting coding as one of the biggest areas of improvement. Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5. On CursorBench 4.0, its best score reached 55.5%, compared with 34.1% for the previous Sonnet model and 57.8% for Opus 5.5.
Anthropic also says the model performs much closer to Opus 5.5 on several knowledge-work evaluations. On GDPval-AA, which measures tasks across multiple occupations and industries, Sonnet 5.5 scored 1,844 compared with 1,846 for Opus 5.5 and 1,449 for Sonnet 5.
The company says early testers also saw improvements in writing, design and collaboration. Anthropic specifically points to document, slide and spreadsheet creation as areas where Sonnet 5.5 can produce more polished results with fewer steps.
The release comes less than a week after Anthropic introduced Opus 5.5. Haiku 5.5, which is intended for higher-volume and more price-sensitive workloads, is expected to join the lineup in the coming weeks.
Anthropic is also giving the new model additional security controls because of its stronger cybersecurity performance. Sonnet 5.5 is the first Sonnet model to launch with cyber safeguards and fallback systems similar to those used for Anthropic’s most capable models.
The company says higher-risk cybersecurity requests can fall back to Sonnet 5, while routine software development remains available. Anthropic has also added protections intended to prevent large-scale model distillation attacks and reasoning extraction.
At the same time, Anthropic says Sonnet 5.5 does not represent a new frontier in overall model capability. Its alignment evaluation therefore focused on risks that can appear across capability levels, including misleading users, acting against user intent and cooperating with high-stakes misuse.
On an automated audit covering roughly 1,850 scenarios, Anthropic says Sonnet 5.5 matched or improved on Sonnet 5 across most measures of alignment, misuse resistance and honesty. The company also says it found no evidence that the model consistently pursued goals that conflicted with user intent.
“Focusing on alignment and safety has been a key part of our mission from the very beginning,” Chu said. “This is something that we’ve always prioritized across our models.”
With Sonnet 5.5, Anthropic is focusing less on pushing maximum capability and more on delivering stronger performance at lower cost and higher speed. The model is designed to sit between Opus 5.5’s higher-end reasoning and the upcoming Haiku 5.5, giving developers another option for everyday coding, knowledge work and production tasks.
This analysis is based on reporting from CNBC & Anthropic.
Image courtesy of Anthropic.
This article was generated with AI assistance and reviewed for accuracy and quality.
