The short version
- This is a generation update, not a price cut. $2 input / $10 output is the same as the previous-generation Claude Sonnet 5. Unlike Opus 5.5, this model did not get cheaper than its predecessor.
- On paper, the specs match Opus 5.5 (our comparison). 1M context, 128K max output, knowledge cutoff June 2026 — the specs match the tier above it.
- Thinking is "Adaptive," on by default, with a default effort of
high. Unlike Opus 5.5's "Adaptive (always on)," Sonnet 5.5 documents a lowest setting,between_tools, that turns off up-front thinking. - No benchmarks published. The official Sonnet 5.5 overview page carries no benchmark table, unlike the Opus 5.5 announcement. We do not invent numbers — we say "not published."
Pricing (USD per 1M tokens, from official docs)
Verified on the Sonnet 5.5 overview page and the official pricing documentation. Standard rates, matching our pricing comparison.
| Item | Claude Sonnet 5.5 | Claude Sonnet 5 (previous gen) | Claude Opus 5.5 (tier above) |
|---|---|---|---|
| Input | $2.00 | $2.00 | $4.00 |
| Output | $10.00 | $10.00 | $20.00 |
| Cache read (cache hit) | $0.20 | $0.20 | $0.20 |
| Cache write (5-minute TTL) | $2.50 | $2.50 | $5.00 |
| Cache write (1-hour TTL) | $4.00 | — | — |
※ Cache prices for Sonnet 5 and Opus 5.5 match the official figures listed in our pricing comparison. The $4 one-hour cache write for Sonnet 5.5 is stated on its overview page; Sonnet 5's equivalent is not published on our site, so it is shown as "—".
The general cache rules Anthropic publishes
“5-minute cache write | 1.25x base input price … 1-hour cache write | 2x base input price … Cache read (hit) | 0.1x base input price (0.025x on Claude Fable 5.1 and Claude Mythos 5.1; 0.05x on Claude Opus 5.5)”
— Claude Platform Docs, "Pricing" (verified 2026-10-01)
- Sonnet 5.5 uses the standard cache discount (10% of input), but the absolute number is low: $0.20 per 1M cached tokens — cheaper in cash terms than Fable 5.1's $0.25, even though Fable 5.1 gets the deeper 2.5% multiplier. Discount depth and absolute cost are different things.
- Batch API is a 50% discount on input and output, per the official pages.
- US-only inference costs 1.1x. Verbatim: “For Claude 4.6 and later models, specifying US-only inference through the
inference_geoparameter incurs a 1.1x multiplier on all token pricing categories, including input tokens, output tokens, cache writes, and cache reads.” - Partner platforms set their own regional prices. Verbatim: “Partner-operated platforms (Bedrock and Google Cloud) have independent regional pricing.”
Primary sources: Sonnet 5.5 overview (pricing block, verified 2026-10-01) · Official pricing docs (cache multipliers, inference_geo, partner pricing)
Chart: Where Sonnet 5.5 sits on unit price
← scroll horizontally →
Specifications (from the official docs)
Released
2026-09-28
Context
1M tokens
Max output
128K tokens
Knowledge cutoff
Jun 2026
Thinking
Adaptive (on by default)
Default effort
high
Status
Active (latest)
Model ID (Claude API)
claude-sonnet-5-5
The official overview table is a single row (verbatim).
“| Claude Sonnet 5.5 | 1M | 128K | $2 / $10 | Fast | Adaptive | high | Jun 2026 |”
— columns are Context / Max output / Price per MTok / Latency / Thinking / Default effort / Knowledge cutoff (verified 2026-10-01)
Source: Claude Platform Docs, "Claude Sonnet 5.5" (Released September 28, 2026 / Context 1M / Max output 128K / Reliable knowledge cutoff Jun 2026 / Training data cutoff Jun 2026 / Availability: Active (latest))
Where it runs, and the model ID on each platform
Same model, different call strings. These are quoted straight from the official page.
| Platform | Model ID |
|---|---|
| Claude API | claude-sonnet-5-5 |
| Amazon Bedrock | anthropic.claude-sonnet-5-5 |
| Google Cloud (Vertex AI) | claude-sonnet-5-5 |
| Microsoft Foundry | claude-sonnet-5-5 |
| Claude Platform on AWS | claude-sonnet-5-5 |
⚠️ Watch the prefix: only Amazon Bedrock adds
anthropic.. More importantly, data handling differs by platform — retention, training and location are governed by the platform you call, not by the model. We break this down in Glossary #8: The four data requirements and #9: Zero operator access.
Thinking — documented differently from Opus 5.5
The Sonnet 5.5 overview page says this, verbatim.
“Adaptive thinking is on by default. The lowest thinking setting is
between_tools, which turns off up-front thinking. It works at high effort or below.”
— Claude Platform Docs, "Claude Sonnet 5.5" (verified 2026-10-01)
| Item | Claude Sonnet 5.5 | Claude Opus 5.5 |
|---|---|---|
| Thinking entry | Adaptive | Adaptive (always on) |
| Default effort | high | medium |
| Can up-front thinking be turned off? | Yes — the lowest setting between_tools turns off up-front thinking, at high effort or below | No — see "thinking can't be disabled" in our Opus 5.5 deep dive |
🔍 Why it matters: both models say "Adaptive," but only Sonnet 5.5 documents an escape hatch to spend fewer thinking tokens. For high-volume, low-complexity work, that difference lands directly on your output-token bill — so "Sonnet is cheaper" is not only about the unit price.
What we could not verify (no estimated numbers)
Our rule: anything not confirmable in a primary source is marked "not verified." We do not publish guessed figures.
| Item | Status | Notes |
|---|---|---|
| Benchmark scores | Not published | The Sonnet 5.5 overview page has no benchmark table (the Opus 5.5 announcement had nine rows). |
| List of available regions | Not verified | The overview page only states "Availability: Status = Active (latest)". What is documented: the 1.1x US-only inference multiplier via inference_geo, and that Bedrock / Google Cloud have independent regional pricing. |
| Rate limits and tiers | Not published | Not mentioned on the overview page. |
| Introductory or limited-time pricing | Not published | No such note exists, so $2 / $10 is treated as the standard price (same as Sonnet 5). |
| ZDR / ZOA availability for this model | Not verified | Per-model retention conditions (e.g. Covered Models) are not stated on the model overview page. See our ZOA explainer for where each vendor documents this. |
How it relates to other pages on this site
| Model | On this site | Relationship |
|---|---|---|
| Claude Opus 5.5 | Dedicated page | Tier above, twice the price ($4 / $20, default effort medium, thinking always on). Sonnet 5.5 sits below it. |
| Claude Sonnet 5 | Listed in Pricing | Previous generation, identical $2 / $10. The story is "same price, newer generation," not "cheaper." |
| Claude Fable 5.1 / Mythos 5.1 | Dedicated pages | Flagship tier ($10 / $50). Sonnet 5.5 is the general-purpose tier below them. |
| Claude Haiku 5.5 | — | The other model the Opus 5.5 announcement said would follow "in the coming weeks." As of 2026-10-01 this site has no page for it, and we did not verify its availability in this update. |
Sources: Anthropic, "Introducing Claude Opus 5.5" (Sonnet 5.5 / Haiku 5.5 "will follow in the coming weeks") · Sonnet 5.5 overview
Good fit / watch out
- High-volume work on a budget: $2 input and $0.20 cache reads are on the low side for a 1M-context model (compared with the models listed in our price comparison), and long reused contexts are exactly where that matters.
- Workloads that need to cap thinking: dropping to
between_toolssaves output tokens on classification and extraction — something Opus 5.5 cannot do. - Migrating from Sonnet 5: the unit price is unchanged, so the decision rests on quality and total tokens. Since Sonnet 5.5's benchmarks are not published, run your own evaluation set instead of comparing numbers.
- Anything with audit or contract implications: retention, operator access, training terms and data location are not on the model page. Start from ZOA and the four requirements.
- Latency-sensitive work: the official table lists Latency as Fast (Opus 5.5 is Moderate). Concrete latency or throughput figures are not published.
Estimate your real cost
Compare token costs across major models, including Claude Sonnet 5.5.
🧮 Open the token calculator →📊 Full pricing comparison · 🤖 Opus 5.5 deep dive · 🚀 Fable 5.1 · 🔐 Zero operator access · 🗓️ Release timeline
⚠️ Disclaimer
- Pricing and specifications were verified against Anthropic's official documentation (Claude Platform Docs "Sonnet 5.5 overview" and "Pricing") on 2026-10-01. They can change without notice — always confirm on the official pages.
- This page publishes no benchmark scores, because none are published officially. We do not estimate or copy numbers from elsewhere.
- Items marked "not verified" (regions, rate limits, limited-time pricing, per-model data retention) will be updated once primary sources are available.
- This site is informational only and is not affiliated with any provider.