Claude Sonnet 5.5 Is Faster, Cheaper, and Strong on Agentic Coding
Anthropic released Claude Sonnet 5.5 on September 28, 2026, the newest version of its mid-tier Sonnet model. The company says the model is 30 percent faster than its predecessor at significantly lower cost. TechCrunch
Anthropic describes Sonnet 5.5 as an assistant for everyday tasks, including coding and drafting office documents. The company says Sonnet 5.5 costs up to 30% less than Sonnet 5 for most work. That discount adds to Sonnet 5 pricing of $2 per million input tokens and $10 per million output tokens, tokens being the small chunks of text the model reads and writes, which Anthropic has now made permanent. TechCrunch
Anthropic also reports benchmark scores for agentic coding, in which the model handles multi-step coding work by calling tools such as editors and test runners. On Anthropic's own benchmarks, Sonnet 5.5 scored ahead of Opus 5.5 on those workloads. Opus is the larger, more expensive tier in the lineup.
Anthropic says Sonnet 5.5 has cyber capabilities comparable to Opus 5. As a result, Sonnet 5.5 is the first Sonnet model subject to the same cyber safeguards that apply to Fable and Opus. The company has not detailed new controls in the materials cited here, only the change in scope.
Sonnet 5.5 follows Opus 5.5 by less than a week. Anthropic released Claude Opus 5.5 on September 22 as the first model in the Claude 5.5 family, with output tokens priced at $20 per million tokens. TechCrunch At that launch, Anthropic said Claude Sonnet 5.5 and Claude Haiku 5.5 would arrive in the coming weeks. BNN Bloomberg Sonnet 5.5 has now shipped. Anthropic plans to release a new version of Haiku, its smallest model, in the coming weeks.
The broader context here is cost and speed engineering rather than a straight race for higher scores. In agentic deployments, token volume multiplies quickly across planning, retrieval, execution and verification loops. It is similar to cutting both fuel cost and travel time for a delivery fleet that makes many trips per order. A 30 percent faster response combined with up to 30 percent lower cost per task compounds across those loops.
In my view, engineering leads should test the benchmark ordering on their own work. Vendor benchmarks require independent validation. Still, the direction fits how many teams already operate. They reserve the largest model for planning or difficult edge cases and send routine implementation, refactoring and document generation to a smaller, faster model. If a Sonnet-class model can reliably run the agentic coding loop, default routing will shift.
Looking at what this means for deployment, the safeguards decision matters as much as the price. Extending Fable and Opus-level cyber safeguards to Sonnet suggests Anthropic expects wide and capable use at this tier. For enterprise users, one tier inherits the stricter regime, which simplifies policy but puts more weight on evaluation. Teams will need to check whether speed and cost gains hold on their own repositories, toolchains and access controls. With Opus 5.5 and Sonnet 5.5 out and Haiku 5.5 still to come, builders gain a near-term option to standardize on one family while tuning spend per request, which allows more code reviewed, more documents drafted and more tests run at lower delay and lower bill.
