Anthropic has released Claude Sonnet 5.5, its second model launch in the Claude 5.5 family ahead of a planned IPO. While base token pricing remains unchanged at $2 per million input and $10 per million output tokens, the model is 30% faster and reduces per-task costs by up to 30% due to token and tool-call efficiencies. Sonnet 5.5 shows massive coding benchmark gains and nearly matches the flagship Opus 5.5 on knowledge tasks. The launch comes amid safety scrutiny and shortly after CEO Dario Amodei called for an industry slowdown.
Releases in the Claude 5.5 family
- ▪Anthropic announced that Claude Haiku 5.5, a model designed for high-throughput, low-cost use cases, will be released in the coming weeks.
- ▪Anthropic released Claude Sonnet 5.5 on September 28, 2026, marking the second model launch in its Claude 5.5 family.
Speed and processing improvements
- ▪Zendesk tested Claude Sonnet 5.5 across hundreds of customer support cases and found that tickets were processed 20 percent faster.
- ▪Claude Sonnet 5.5 generates output more than 30 percent faster than its predecessor, Claude Sonnet 5.
Cost reduction per task
- ▪At low or medium effort settings, Claude Sonnet 5.5 can surpass Claude Sonnet 5's best scores at approximately one-tenth the cost per task.
- ▪Claude Sonnet 5.5 reduces the total cost of completing a task by up to 30 percent because it uses fewer tokens and fewer tool calls.
API pricing details
- ▪Anthropic kept the API pricing for Claude Sonnet 5.5 identical to Sonnet 5 at $2 per million input tokens and $10 per million output tokens.
- ▪Claude Sonnet 5.5 offers a discount rate of $0.20 per million cache-read tokens and $2.50 per million cache-write tokens.
Benchmark evaluation results
- ▪Claude Sonnet 5.5 scored 1,844 points on the GDPval-AA knowledge-work benchmark, nearly matching Claude Opus 5.5's score of 1,846.
- ▪Claude Sonnet 5.5 scored 70.6 percent on Terminal-Bench 4.0, an agentic coding evaluation, compared to Claude Sonnet 5's score of 10.3 percent.
- ▪On CursorBench 4.0, Claude Sonnet 5.5 scored 55.5 percent, landing just over two points below Claude Opus 5.5's score of 57.8 percent.
Platform availability and privacy
- ▪Claude Sonnet 5.5 is available across major cloud platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.
- ▪Anthropic offers Claude Sonnet 5.5 with zero-data-retention support across all of its available platforms.
Safety and security safeguards
- ▪Anthropic introduced safety classifiers in Claude Sonnet 5.5 designed to block model distillation attacks that attempt to extract its reasoning.
- ▪Anthropic added cybersecurity safeguards to Claude Sonnet 5.5, automatically rerouting high-risk cybersecurity requests to Claude Sonnet 5.
Calls for AI safeguards
- ▪Anthropic released Claude Sonnet 5.5 shortly after CEO Dario Amodei publicly urged AI companies to slow down the development of advanced models.
- ▪Microsoft co-founder Bill Gates called for legislation in Congress on September 27, 2026, arguing that artificial intelligence self-regulation is insufficient.
Debatable claims
- ▪Anthropic's rapid model releases undermine its CEO's calls for an AI slowdown
- ▪Mid-tier AI models are capable enough to replace premium frontier models for enterprise tasks
- ▪A voluntary slowdown of frontier AI development is impossible to achieve
Story comments
Loading comments…