The short answer
Claude Sonnet 5.5 is a straight efficiency upgrade: same $2 / $10 price as Sonnet 5, but more than 30% faster output and up to 30% lower cost per task, with agentic-coding scores that Anthropic puts on par with Opus 5.5. For teams that build on Claude via the Anthropic API, AWS, Google Cloud or Azure, the new model ID is claude-sonnet-5-5.
Two things deserve engineering attention. First, the savings come from the model using fewer tokens and tool calls, so they show up in cost per completed task, not on the price sheet — you only see them if you measure that way. Second, Sonnet 5.5 is the first Sonnet with cyber safeguards: higher-risk security requests are answered by Sonnet 5 instead. In an AI agent pipeline, that means two different models can serve the same workflow.
What changed in Sonnet 5.5?
The price did not move; the work per dollar did. Anthropic kept Sonnet list pricing at $2 per million input tokens and $10 per million output tokens, and says the new model finishes the same tasks with fewer tokens and fewer tool calls. Combined with output that is more than 30% faster, that is where the “up to 30% cheaper per task” figure comes from. Early customer numbers cited by VentureBeat point the same way: Lovable reported one-third fewer tool calls and roughly half as many shell executions, and Box measured 2.4× faster responses with 12% fewer tokens.
The capability jump is largest in agentic coding. Anthropic reports 70.6% on Terminal-Bench 4.0, up from 10.3% for Sonnet 5 and ahead of the 66.4% it published for Opus 5.5 a week earlier. On broader work benchmarks the two models are almost level: 1844 versus 1846 on GDPval-AA and 80.1% versus 81.8% on OSWorld 2.1. Anthropic also says Sonnet 5.5 at Low or Medium effort can beat Sonnet 5’s best score at roughly one-tenth of the cost per task. All of these are vendor-run results — a reason to test, not a verdict.
TechCrunch describes the model as Anthropic’s everyday work partner, positioned for well-scoped tasks such as fixing bugs and producing documents, slides and spreadsheets, while Opus stays the tier for the hardest problems. Sonnet 5.5 is available on the Claude Platform and through AWS, Google Cloud and Microsoft Azure, with zero data retention support.
How do the new cyber safeguards work?
Sonnet 5.5 is the first Sonnet model to ship with the cybersecurity safeguards Anthropic previously reserved for its top models. TechCrunch reports that its cyber capabilities are comparable to Opus 5, which is why the same class of protections now applies. When a request is classified as higher-risk cybersecurity work, it is served by Sonnet 5 instead of Sonnet 5.5. Organizations doing legitimate defensive work can apply to Anthropic’s Cyber Verification Program for tiered access to full capability.
For most product teams this will rarely trigger. For teams building security scanners, SOC copilots or penetration-testing assistants on Claude, it matters: the same workflow may now be answered by two different models, with different quality and different behavior. If your integration follows the pattern Anthropic documented for Opus 5.5, check how fallbacks are configured on the API and what a flagged call returns before you rely on it in an unattended agent.
What it means for US & EU software teams
First, the default tier shifts down. If Sonnet 5.5 really performs within a couple of points of Opus 5.5 on your workloads at half the list price, many agents that were moved to Opus for quality can move back. That is a routing decision, not a global switch: keep Opus 5.5 for the long, ambiguous tasks where it still wins, and send well-scoped coding, triage and document work to Sonnet 5.5.
Second, cost per token is the wrong metric. Because the savings come from fewer tokens and tool calls, a dashboard that tracks price per million tokens will show no change. Track cost and latency per completed task instead, per workflow and per effort level. The five effort levels are now the main cost lever: a support-triage agent at Low effort and a refactoring agent at Xhigh are effectively different products.
Third, model inventories need to reflect fallbacks. Under the EU AI Act’s documentation duties, GDPR accountability and most enterprise vendor-risk reviews, you are expected to know which system produced an output. With cyber fallbacks, a system documented as running on Sonnet 5.5 may serve some answers from Sonnet 5. Log the model reported on each response, not just the model requested, and keep that record next to your evaluation results.
What to do before you switch
- Re-run your eval set. Compare Sonnet 5, Sonnet 5.5 and Opus 5.5 on your own tasks, including output length and style, before moving traffic.
- Measure cost per completed task. Record tokens, tool calls and latency per task; that is where the claimed 30% shows up, or does not.
- Pick effort levels per workflow. Start cheap-path workflows at Low or Medium and raise effort only where quality demands it.
- Handle fallbacks explicitly. Log which model answered, alert on unexpected fallbacks or safety stops, and apply to the Cyber Verification Program if your product is security tooling.
- Keep a rollback. Pin
claude-sonnet-5as a fallback route until your own metrics confirm the upgrade.
Frequently asked questions
What did Anthropic release on September 28, 2026?
Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family after Opus 5.5 (September 22, 2026). It is available through the Claude Platform as claude-sonnet-5-5 and on Amazon Web Services, Google Cloud and Microsoft Azure, with zero data retention available. Anthropic says it generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task. Haiku 5.5 is expected in the coming weeks.
How much does Claude Sonnet 5.5 cost?
List pricing is unchanged from Sonnet 5: $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 and cache writes at $2.50 per million tokens. The savings come from efficiency, not a price cut: Anthropic says the model completes tasks with fewer tokens and fewer tool calls. For comparison, Opus 5.5 is listed at $4 and $20 per million tokens.
Is Sonnet 5.5 good enough to replace Opus 5.5 for coding agents?
On Anthropic's published numbers it is close. Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0 against 66.4% for Opus 5.5, 1844 versus 1846 on GDPval-AA and 80.1% versus 81.8% on OSWorld 2.1, at half the list price. These are vendor-run benchmarks, so run both models on your own task set before moving production traffic; for many well-scoped coding and document tasks Sonnet 5.5 is likely to be the cheaper default.
What are the new cybersecurity safeguards in Sonnet 5.5?
Sonnet 5.5 is the first Sonnet model to launch with the cyber safeguards Anthropic uses on its most capable models. Higher-risk cybersecurity requests fall back to Sonnet 5 instead of being answered by Sonnet 5.5. Organizations doing defensive security work can apply to Anthropic's Cyber Verification Program for tiered access. Teams building security tooling on Claude should test which of their prompts trigger the fallback.
What should engineering teams do before switching to Sonnet 5.5?
Treat it as a model change: re-run your evaluation set, compare cost per completed task rather than price per token, decide which effort level each workflow needs, log which model actually answered each request, and handle safety fallbacks explicitly in agent loops. Keep Sonnet 5 pinned as a rollback option until your own metrics confirm the gain.
Sources
Anthropic — Introducing Claude Sonnet 5.5 (company announcement)
Anthropic — Claude Sonnet 5.5 System Card (September 28, 2026)
TechCrunch — Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner
CNBC — Anthropic Sonnet 5.5 launch: price, features and safety
VentureBeat — Anthropic launches Claude Sonnet 5.5 with 30% cost reduction per task