Quick Facts
- Claude Sonnet 5.5 launched Sept. 28, 2026, six days after Opus 5.5, with the same per-token price as its predecessor.
- Per-task costs drop up to 30% because the model completes work using fewer tokens and fewer tool calls, not cheaper pricing.
- Early customers including Slack, Box, Zendesk, and Epic Games reported measurable speed and efficiency gains in production tests.
Anthropic released Claude Sonnet 5.5 on Sept. 28, 2026, the second model in its Claude 5.5 family. The model targets well-scoped everyday tasks: fixing bugs, drafting documents, building slides, and processing structured data.
The headline claim is a 30% reduction in the cost of completing a task. That number does not reflect a price cut. Anthropic kept the rate at $2 per million input tokens and $10 per million output tokens, identical to Sonnet 5. The savings come from the model reaching the same output in fewer steps.
Efficiency Over Price Cuts
Anthropic says Sonnet 5.5 generates output more than 30% faster than Sonnet 5. The model is also designed to reach task completion with fewer tool calls and fewer tokens consumed per session.
Sonnet 5 moves to legacy status but stays available. Its $2/$10 pricing, originally described as introductory, is now permanent. A planned increase to $3/$15, scheduled for Sept. 1, was cancelled.
What Benchmarks Show
On Terminal-Bench 4.0, a test of agentic coding using terminal-based tools and multi-step workflows, Sonnet 5.5 scores 70.6%. Sonnet 5 scored 10.3% on the same test. The new model also outscores Opus 5.5, which posted 66.4%.
On GDPval-AA, which measures performance across 44 occupations, Sonnet 5.5 scores 1,844 against Opus 5.5's 1,846. Sonnet 5 scored 1,449. Sonnet 5.5 also beats Opus 5.5 on AutomationBench (44.7% vs. 42.5%) and HealthBench Professional (69.2% vs. 65.6%). It trails on deep reasoning and long-context tasks, falling 7.5 points behind Opus 5.5 on Humanity's Last Exam without tools.
Customer Results
Slack Principal Engineer Curtis Allen said Sonnet 5.5 outperformed Sonnet 5 on nearly all offline Slackbot evaluations without prompt changes, producing roughly 14% fewer output tokens per session.
Lovable co-founder and CTO Fabian Hedin said the model required about one-third fewer tool calls and roughly half as many shell executions to complete coding tasks. Atlassian reported that Rovo Agents ran actions up to 30% faster compared to Sonnet 5.
Box reported 2.4x the processing speed and 12% fewer total tokens. Zendesk saw support tickets processed 20% faster. Base44 AI Engineering Lead Gabriel Grinberg evaluated the model across 118 app builds and found Sonnet 5.5 matched Opus 5 quality in an average of 3.6 iterations per build, compared to 7.7 for Opus 5.
Balyasny Asset Management ran Sonnet 5.5 against 2,441 financial tasks. The model averaged roughly 121,000 tokens per task. Sonnet 5 averaged about 497,000 on the same tasks.
Safety Architecture
Sonnet 5.5 ships with the same cybersecurity safeguards as Opus 5.5, a first for a mid-tier Anthropic model. That includes classifiers designed to block reasoning extraction through distillation attacks.
Claude Haiku 5.5, targeted at high-volume and cost-sensitive applications, is expected to join the family in the coming weeks.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.