Claude Sonnet 5.5 Is Here: Faster, Cheaper Agents for Automation Teams

A Cheaper, Faster Model Just Landed for Agent Builders

On September 28, 2026, Anthropic released Claude Sonnet 5.5, the mid-tier model that many agencies use as their default for automation and coding agents. The pitch is simple: more speed, fewer tokens per task, and the same per-token price as Sonnet 5. For anyone running agent workflows at volume, that combination goes straight to the margin line.

What Anthropic Announced

According to Anthropic's announcement and TechCrunch's coverage, Sonnet 5.5 is a significant step up from Sonnet 5 (released in June 2026):

• Speed: more than 30% faster than Sonnet 5.
• Efficiency: up to 30% fewer tokens per task, which lowers real cost even though list pricing did not change.
• Context: a 1M-token context window.
• Model ID: claude-sonnet-5-5, available on Claude.ai and the Claude Platform, and through AWS, Google Cloud, and Microsoft Azure.

Pricing at a Glance

List pricing is $2 per million input tokens and $10 per million output tokens. Cache reads are $0.20 per million and cache writes are $2.50 per million. Because the model needs fewer tokens to finish the same job, your effective cost per completed task should drop even with identical rates. Test this on your own workloads rather than assuming it, since savings depend on the task.

Benchmarks Worth Noting

Anthropic reports 70.6% on Terminal-Bench 4.0 (agentic coding), ahead of the 66.4% it lists for the more expensive Opus 5.5. It also reports 80.1% on OSWorld (computer use) and near-parity with Opus 5.5 on the GDPval-AA knowledge-work evaluation. These are vendor-reported numbers, so treat them as a starting point for your own evaluations. The direction is still clear: the gap between the mid-tier and flagship models keeps shrinking for practical agent work.

Why It Matters for Agencies and Automation Builders

Most client automations do not need the biggest model. They need reliable tool use, decent speed, and a predictable bill. A faster, more token-efficient Sonnet helps in a few concrete ways:

Safety Changes You Should Know About

Anthropic says Sonnet 5.5 ships with cybersecurity safeguards comparable to those on Opus 5.5, plus biology safeguards and measures against distillation attacks. In practice, some borderline security-related requests may be handled more conservatively than before. If you build security tooling for clients, test your prompts.

How to Try It Safely

Do not swap models in production on launch day. Instead, run your existing prompts and test cases against claude-sonnet-5-5, compare accuracy, latency, and token counts against your current model, and roll it out gradually behind a config flag. Anthropic has also said an updated Haiku model is planned in the coming weeks, so keep your model choice configurable.

Key Takeaways

• Sonnet 5.5 is 30%+ faster and uses up to 30% fewer tokens than Sonnet 5, at the same $2/$10 list price.
• It offers a 1M-token context window and is available on the Claude Platform, AWS, Google Cloud, and Azure.
• Vendor benchmarks show it close to, and on some agentic coding tests ahead of, Opus 5.5.
• Validate on your own workloads and roll out gradually.
• Keep the model name in configuration so you can adopt the upcoming Haiku update quickly.