Anthropic unveils Claude Sonnet 5.5: faster and up to 30% cheaper
XORit Editorial · Sources, AI assistance and corrections
Anthropic unveiled Claude Sonnet 5.5 on September 28, 2026, the second model in the Claude 5.5 family. According to the developer, it runs more than 30% faster than Sonnet 5 and costs up to about 30% less on typical tasks.
Token pricing is unchanged: $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20. The savings come from the model spending noticeably fewer tokens on the same work.
Coding and knowledge
The biggest gain is in agentic coding. On Terminal-Bench 4.0, Sonnet 5.5 scores 70.6% versus 10.3% for Sonnet 5, and on CursorBench 4.0 it gets 55.5% versus 34.1%, trailing Opus 5.5 by only about two points.
In knowledge work, the model has caught up with the higher-tier version: on GDPval-AA it is nearly level with Opus 5.5 and about 400 points above Sonnet 5.
Anthropic adds that benchmarks show only part of the model's capabilities. According to the company's internal tests and feedback from external testers, Opus 5.5 remains stronger on complex open-ended tasks. Sonnet 5.5 is better suited to everyday, well-defined tasks.
Security and availability
Because of its increased cybersecurity capabilities, the model is the first among Sonnet to ship with special safeguards similar to those used for Opus 5.5. Ordinary software development and most life sciences work do not fall under them.
Sonnet 5.5 is already available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Another model in the family, Claude Haiku 5.5 for high-volume and price-sensitive scenarios, is promised in the coming weeks.
Translation editor: Салтанов А.
Source: www.anthropic.com ↗