Claude Sonnet 5.5 is Anthropic’s September 28, 2026 mid-tier model — the second member of the Claude 5.5 family after Claude Opus 5.5. It keeps Sonnet 5’s $2 / $10 price, runs 30%+ faster, and needs far fewer tokens per task, which Anthropic estimates cuts cost per task by up to 30%.
Where Opus 5.5 is built for complex work requiring careful judgment, Sonnet 5.5 is positioned as the strongest model for well-scoped everyday tasks, bug fixing, and polished documents, slides and spreadsheets — with, testers noted, a sharp eye for design.
Benchmarks
Vendor-reported at launch (Anthropic, September 28, 2026). GPT-6 Sol is shown where OpenAI published a comparable number; Anthropic notes that Terminal-Bench and OpenAI did not publish a GPT-6 Sol score, so GPT-5.6 Sol is used in its cost charts instead.
| Benchmark | What it measures | Sonnet 5.5 | Sonnet 5 | Opus 5.5 | GPT-6 Sol |
|---|---|---|---|---|---|
| Terminal-Bench 4.0 | Agentic terminal coding | 70.6% | 10.3% | 66.4% | — |
| CursorBench 4.0 | Real Cursor coding sessions | 55.5% | 34.1% | 57.8% | — |
| GDPval-AA v2.1 | Real-world knowledge work (Elo) | 1844 | 1449 | 1846 | 1487 |
| AA-Briefcase v1.1 | Professional deliverables (Elo) | 1811 | 1359 | 1822 | 1483 |
| Humanity’s Last Exam | Reasoning (with tools) | 64.5% | 54.9% | 67.7% | — |
| OSWorld 2.1 | Computer use (partial credit) | 80.1% | 57.0% | 81.8% | — |
| Chartography | Chart recognition (no tools) | 61.6% | 15.6% | 64.4% | 53.6% |
Two patterns stand out. First, Sonnet 5.5 leaps over Sonnet 5 almost everywhere — Terminal-Bench 4.0 goes from 10.3% to 70.6% and Chartography from 15.6% to 61.6%. Second, effort settings matter more than the model name: at Low or Medium effort Sonnet 5.5 beats Sonnet 5’s best score on several benchmarks for roughly a tenth of the cost per task, while at Max it can match Opus 5.5 at a similar cost. Anthropic’s own caveat: in its testing, Opus 5.5 still wins on open-ended work that demands sustained judgment.
Pricing
| Prices per 1M tokens | Claude Sonnet 5.5 | Claude Opus 5.5 |
|---|---|---|
| Input | $2 | $4 |
| Output | $10 | $20 |
| Cache reads | $0.10 (cut from $0.20 on Oct 7) | $0.20 |
| Cache writes | $2.50 | $5 |
Prompt caching saves up to 90% and batch processing 50%. For workloads that must run in the US, US-only inference is 1.1x. The October 7 cache-read cut means Sonnet 5.5 now runs around 20% cheaper on most agentic work.
Specs
| Spec | Value |
|---|---|
| API model ID | claude-sonnet-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Thinking | Adaptive; default effort Medium in Claude Code and apps, High on the Claude Platform |
| Reliable knowledge cutoff | June 2026 |
| Modalities | Text and image in, text out |
| Platforms | Claude.ai, Claude Platform, Amazon Web Services, Google Cloud, Microsoft Foundry |
Safety and safeguards
- Alignment. On Anthropic’s roughly 1,850-scenario automated behavioral audit it improves on or matches Sonnet 5 on most measures, and on containment evaluations it comes close to Opus 5.5 — the least likely of any Claude model to probe the limits of its container.
- Cybersecurity. Because its cyber capability is a large improvement over Sonnet 5, it is the first Sonnet to launch with cyber safeguards and fallbacks like Opus 5.5’s. Routine bug-finding and fixing is unaffected; higher-risk tasks visibly fall back to Sonnet 5. Defenders can apply to the expanded Cyber Verification Program.
- Biology. Same safeguards as Sonnet 5; organizations can apply to the Life Sciences Verification Program.
- Distillation. First Sonnet with classifiers that prevent reasoning extraction, and it expands preserved thinking, so thinking cannot be decoupled from the account that produced it.
How Sonnet 5.5 compares
- vs Claude Opus 5.5: half the price ($2 / $10 vs $4 / $20) and faster, close on CursorBench and GDPval-AA. Choose Opus for sprawling, open-ended jobs; Sonnet for scoped work and as an orchestrated subagent.
- vs Claude Haiku 5.5: Haiku costs a fifth as much (from $0.10 / $0.50) but trails on Terminal-Bench 4.0 (39.2% vs 70.6%) — use it for classification, extraction and routing.
- vs Claude Sonnet 5: a drop-in upgrade at the same list price with a large capability jump.
- vs GPT-6 Sol / 6.1 Sol: the closest OpenAI equivalents sit at the same $2 / $10 list price.
Limitations
- Not the judgment ceiling. Anthropic’s testers still prefer Opus 5.5 for complex, ambiguous projects.
- Safeguard fallbacks. Higher-risk cyber requests fall back to an older model, so security-adjacent agents can behave differently from the headline numbers.
- Restated figures. Later Anthropic pages show slightly different Sonnet 5.5 numbers on some benchmarks (for example GDPval-AA 1840), so treat decimals cautiously.
For the wider picture, see the 2026 LLM guide.
Best for
- Well-scoped everyday coding — bug fixing, reviews, refactors — at Sonnet speed
- Polished documents, slides and spreadsheets
- Agent subagents beneath an Opus 5.5 orchestrator at lower effort
- Design-sensitive front-end work, which testers singled out
Pros & cons
Strengths
- Terminal-Bench 4.0 70.6% (vs Sonnet 5's 10.3%) and CursorBench 4.0 55.5%, within about two points of Opus 5.5
- Same $2 / $10 price as Sonnet 5 but uses far fewer tokens — Anthropic estimates up to 30% less per task — and generates output 30%+ faster
- Cache reads halved to $0.10 per 1M on October 7, so agentic workloads run roughly 20% cheaper again
- GDPval-AA v2.1 of 1844 is only two points below Opus 5.5; first Sonnet to beat Pokémon Red from screenshots alone
Limitations
- Opus 5.5 remains clearly stronger on complex, open-ended work that needs sustained judgment, per Anthropic's own testing
- Higher-risk cybersecurity tasks visibly fall back to Sonnet 5 under the new safeguards
- Some microbiology and virology requests may be flagged in error by the biology safeguards inherited from Sonnet 5
- Benchmark figures are vendor-reported and several were later restated, so expect small differences between Anthropic's pages
Frequently asked questions
When was Claude Sonnet 5.5 released?
September 28, 2026, six days after Claude Opus 5.5. Claude Haiku 5.5 completed the Claude 5.5 family on October 7.
How much does Claude Sonnet 5.5 cost?
$2 per 1M input tokens and $10 per 1M output tokens, the same list price as Sonnet 5. Cache writes are $2.50 per 1M and cache reads were halved to $0.10 per 1M with the Haiku 5.5 launch. Anthropic says it costs up to 30% less per task than Sonnet 5 because it needs fewer tokens.
Is Sonnet 5.5 better than Opus 5.5 for coding?
On Anthropic's table it is close — CursorBench 4.0 55.5% vs 57.8%, GDPval-AA 1844 vs 1846 — and its Terminal-Bench 4.0 score of 70.6% is above Opus 5.5's 66.4%. Anthropic still says Opus 5.5 is clearly stronger on complex, open-ended work that needs sustained judgment, and that Sonnet 5.5 works best on well-scoped tasks.
What is the Claude API model ID for Sonnet 5.5?
`claude-sonnet-5-5`, with a 1M-token context window, up to 128K output tokens and adaptive thinking. Default effort is Medium in Claude Code and the apps and High on the Claude Platform.
Where can I use Sonnet 5.5?
On Claude.ai (web, iOS, Android), the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Foundry.