Anthropic Launches Claude Sonnet 5.5 With New Cyber Safeguards
The everyday-work model keeps Sonnet 5 token prices and launches with cybersecurity fallbacks as Anthropic claims faster output and lower per-task cost.
Edited by Tyronne Panaino
Anthropic launched Claude Sonnet 5.5 on September 28 across its own platform and Amazon Web Services, Google Cloud and Microsoft Azure. The official announcement positions the model as the everyday-work member of the Claude 5.5 family, aimed at well-scoped tasks such as bug fixes and producing polished documents, slides and spreadsheets.
The release matters to developers and enterprise teams choosing between speed, cost and task complexity. Anthropic keeps the same headline token prices as Sonnet 5 while claiming faster generation and lower per-task consumption. It also makes Sonnet 5.5 the first Sonnet release to carry cyber safeguards and fallbacks similar to those used on the company's more capable models.
The model targets routine work below the Opus tier
Anthropic describes Sonnet 5.5 as a lower-cost, faster complement to Claude Opus 5.5 rather than a replacement for its highest-judgment model. The company still positions Opus for complex, open-ended work requiring sustained judgment, while Sonnet is meant to handle more bounded jobs and quicker iteration.
That distinction is important for buyers. A model can be attractive for a high-volume workflow without being the best option for every difficult task. Anthropic itself cautions that benchmark scores capture only one part of capability, so a procurement decision should not reduce to a single evaluation table or vendor ranking.
Token prices stay fixed while task-cost claims change
Sonnet 5.5 keeps Sonnet 5's listed prices: $2 per million input tokens, $10 per million output tokens and $0.20 per million cache-read tokens. Anthropic says the new model usually needs fewer tokens to complete the same work and can cost up to 30% less per task in its testing. It also reports output generation more than 30% faster than Sonnet 5.
Those figures are first-party measurements, not a workload-neutral guarantee. Total application cost can also include tool calls, retries, caching behavior, infrastructure and human review. Teams considering a migration should replay representative tasks and measure completion quality, latency, token use and total cost together rather than assuming the published maximum applies to every prompt.
Effort settings shape the speed-quality tradeoff
Anthropic says Claude Code and its user-facing apps default Sonnet 5.5 to Medium effort, while the Claude Platform defaults to High. Lower settings are intended to answer faster and use fewer tokens; higher settings allow more reasoning and checking.
This gives operators another variable to test. Comparing Sonnet 5 with Sonnet 5.5 at different defaults can blur whether an observed change comes from the model or the effort setting. A controlled evaluation should therefore record the effort level alongside prompts, tools, time and token use.
Cyber fallbacks expand the Sonnet safety layer
Anthropic says Sonnet 5.5 is the first Sonnet model launched with cybersecurity safeguards and fallbacks like those used for its most capable models. Higher-risk cyber requests can visibly fall back to Sonnet 5, while routine software development is intended to remain available.
The biology safeguards remain the same as Sonnet 5's. Anthropic also says the release adds classifiers intended to prevent reasoning extraction through industrial-scale distillation attacks. These are descriptions of the provider's controls; the fetched evidence does not independently test their coverage, false-positive rate or resistance to adaptive attacks.
Availability comes with one migration detail
Developers can select the model on the Claude Platform with the identifier `claude-sonnet-5-5`, and Anthropic lists availability through AWS, Google Cloud and Microsoft Azure. Applications that run Sonnet with thinking turned off need to move to the `between_tools` setting, which keeps up-front thinking disabled while preserving reasoning between tool calls.
The practical checkpoint is a side-by-side workload evaluation followed by a controlled migration. Teams should confirm model availability in their chosen region and provider, preserve their existing test set, pin the effort configuration and verify that fallback behavior is visible in operational logs.
Evidence quality and limits
The release and availability are confirmed by Anthropic's primary announcement. Internal confidence is medium because the speed, efficiency, benchmark and safety results are mainly provider-run, and no independent reproduction was available in the fetched evidence.
Status
Confirmed. Anthropic has launched the model and documented its prices, availability, effort defaults, migration setting and safeguard design. Performance and control effectiveness remain company claims pending broader independent testing.
Sources
Update note: Last reviewed 2026-09-29. We will revise this post if Anthropic changes availability or independent evaluations clarify performance and safeguard behavior.
Sources
- Anthropic — Claude Sonnet 5.5 — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.