Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioning it as the everyday workhorse in the Claude 5.5 family and the cheaper complement to Opus 5.5.
The headline promise is straightforward: Sonnet 5.5 is faster than Sonnet 5, usually cheaper per finished task, and much stronger on the agentic coding and tool-use patterns that matter in Claude Code, Claude.ai, and API deployments.
What changed
Anthropic says Sonnet 5.5 generates outputs 30%+ faster than Sonnet 5 and costs up to 30% less per task in its internal testing. The token price itself has not changed: $2 per million input tokens, $10 per million output tokens, and $0.20 per million cache-read tokens.
That distinction matters. Sonnet 5.5 is not cheaper because the sticker price dropped; it is cheaper because it tends to finish work with fewer tokens and fewer unnecessary steps. Early customer examples in Anthropic's launch post point to fewer tool calls, fewer shell runs, faster support-ticket processing, and lower output-token use in finance and Slackbot-style workflows.
Anthropic is also drawing a sharper product line across the family. Opus 5.5 remains the model for open-ended work that needs sustained judgment. Sonnet 5.5 is for well-scoped everyday work: fixing bugs, implementing features after a larger plan is set, drafting polished documents, cleaning spreadsheets, creating slides, and iterating quickly with a user.
Haiku 5.5 is still not here. Anthropic says the high-volume, cost-sensitive member of the 5.5 family will arrive in the coming weeks.
The benchmark numbers
The biggest jump is in agentic terminal coding. Anthropic reports Sonnet 5.5 at 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5 and 66.4% for Opus 5.5 at its highest reported effort setting.
| Benchmark | Claude Sonnet 5.5 | Claude Sonnet 5 | Claude Opus 5.5 | GPT-6 Sol |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 10.3% | 66.4% | - |
| FrontierCode 1.1 main | 46.2% Max / 52.1% Xhigh | 42.4% | 54.4% | 49.3% |
| CursorBench 4.0 | 55.5% | 34.1% | 57.8% | - |
| GDPval-AA v2.1 | 1,844 Elo | 1,449 Elo | 1,846 Elo | 1,487 Elo |
| AA-Briefcase v1.1 | 1,811 Elo | 1,359 Elo | 1,822 Elo | 1,483 Elo |
| Humanity's Last Exam | 64.5% with tools | 54.9% with tools | 67.7% with tools | - |
| OSWorld 2.1 | 80.1% partial | 57.0% partial | 81.8% partial | - |
| Chartography | 61.6% no tools | 15.6% no tools | 64.4% no tools | 53.6% no tools |
The pattern is not "Sonnet replaces Opus." It is closer to "Sonnet now covers a lot more of the work that used to require escalating." Sonnet 5.5 gets near Opus 5.5 on knowledge-work and computer-use numbers, beats Sonnet 5 by wide margins in several places, and gives developers a cheaper default before reaching for the top model.
Anthropic includes two caveats worth keeping. First, Opus 5.5 is still stronger for complex, open-ended work. Second, Artificial Analysis ran GDPval-AA and AA-Briefcase on a pre-release Sonnet 5.5 deployment affected by a structured-output bug that Anthropic says has since been fixed, so those scores may slightly understate the final model.
Pricing and model details
The model ID is claude-sonnet-5-5. Anthropic's platform docs list a 1 million-token context window, 128K max output for normal use, and up to 300K output tokens through the Message Batches API extended-output beta.
| Model | Input | Output | Cache read | Cache write |
|---|---|---|---|---|
| Claude Sonnet 5.5 | $2 / 1M tokens | $10 / 1M tokens | $0.20 / 1M tokens | $2.50 / 1M tokens |
| Claude Opus 5.5 | $4 / 1M tokens | $20 / 1M tokens | $0.20 / 1M tokens | $5 / 1M tokens |
Effort settings matter more in this generation. In Claude apps and Claude Code, Anthropic says the default effort is Medium. On the Claude Platform, the default is High. Lower effort should suit routine work; higher effort gives the model more room to reason and check itself.
Availability
Sonnet 5.5 is available now across Anthropic's own platform and the major clouds. Anthropic says developers can use it on the Claude Platform with claude-sonnet-5-5, and AWS says customers can access it through both Amazon Bedrock and Claude Platform on AWS.
AWS is pitching the Bedrock route for teams that want AWS infrastructure, regional data residency, and AWS-managed controls such as Guardrails and Knowledge Bases. Claude Platform on AWS is the more native Anthropic path, surfaced through the AWS Console with AWS billing and authentication.
Anthropic also says Sonnet 5.5 is available with zero data retention, matching the enterprise posture it highlighted for Opus 5.5 and Sonnet 5.
Safety changes
Sonnet 5.5 is the first Sonnet model to launch with the cyber safeguards and fallbacks Anthropic previously reserved for its most capable models. Routine software development should keep working, but higher-risk cybersecurity requests can visibly fall back to Sonnet 5.
The biology safeguards are the same as Sonnet 5's. Anthropic says most research, education, clinical work, and ordinary software work should be unaffected, though some microbiology and virology requests may be flagged.
There is also a distillation-protection angle. Because Sonnet 5.5 is substantially more capable than its predecessor, Anthropic says it is adding classifiers intended to prevent large-scale reasoning extraction. The model also expands preserved thinking, which ties Claude's thinking traces more closely to the account that created them.
Why it matters
For builders, the important thing is not just the benchmark jump. It is the routing implication.
If Sonnet 5.5 can handle routine coding, support, document, analysis, and internal-agent tasks with fewer tool calls and lower per-task cost, teams can reserve Opus 5.5 for architecture, ambiguous judgment calls, and harder long-running work. That makes the default Claude stack cheaper without forcing everyone onto a small model.
The short version: Sonnet 5.5 looks like Anthropic's new practical default. Opus 5.5 is still the premium reasoning model, but Sonnet is now fast and capable enough to carry much more of the everyday workload.