Anthropic Releases Sonnet 5.5 With Lower Cost and Faster Performance
Anthropic released Claude Sonnet 5.5 on September 28, 2026, the newest version of its mid-tier Sonnet model. The company says the model works faster and at significantly less cost than its predecessor, specifying a 30 percent speed improvement. TechCrunch
Anthropic describes Sonnet 5.5 as an assistant for everyday tasks, including coding and creating office documents. The pricing claim is specific. Claude Sonnet 5.5 costs up to 30% less than Sonnet 5 for most work. That discount sits on top of Sonnet 5 introductory pricing of $2 per million input tokens and $10 per million output tokens, which Anthropic has now made permanent. TechCrunch
For practitioners, the more notable claim concerns agentic coding. Anthropic's own benchmarks put Sonnet 5.5 ahead of Opus 5.5 on agentic coding workloads. That matters for production workloads. Opus is the larger, more expensive tier. A mid-tier model leading on multi-step coding tasks with tool use changes how teams allocate models across a pipeline.
The release also changes the safeguards posture for the Sonnet line. Anthropic claims Sonnet 5.5 has comparable cyber capabilities to Opus 5. As a result, Sonnet 5.5 is the first Sonnet model that will be 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 scope change.
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 its 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 be released in the coming weeks. BNN Bloomberg Sonnet 5.5 is now delivered. Anthropic plans to release a new version of Haiku, its smallest model, in the coming weeks.
The broader context here is cost and latency engineering rather than a straight capability race. In agentic deployments, token volume multiplies quickly across planning, retrieval, execution and verification loops. A 30 percent reduction in latency combined with up to 30 percent lower cost per unit of work compounds across those loops. Throughput improves. Iteration gets cheaper.
In my view, the benchmark ordering deserves close attention from engineering leads. Vendor benchmarks require independent validation. Still, the direction is consistent with what many teams already do in practice. They reserve the largest model for planning or difficult edge cases and push routine implementation, refactoring and document generation to a smaller, faster model. If a Sonnet-class model can handle the agentic coding loop reliably, the default routing changes.
Looking at what this means for deployment, the safeguards decision is as important as the price. Extending Fable and Opus-level cyber safeguards to Sonnet suggests Anthropic expects this tier to be used widely enough, and capably enough, to warrant the same controls. For enterprise users, that simplifies policy in one sense. One tier inherits the stricter regime. It also puts more weight on evaluation. Teams will need to test whether the speed and cost gains hold on their own repositories, toolchains and access controls, not only on reference tasks.
The cadence is quick. Opus 5.5, now Sonnet 5.5, with Haiku 5.5 still to come. For builders, that creates a near-term option to standardize on a single model family across complexity levels while tuning spend per request. The practical gain is mundane and valuable. More code reviewed, more documents drafted, more tests run, at lower latency and lower bill.
