Claude Sonnet 5.5 Pricing: API Rates, Plans, and What a Task Really Costs

‍

Sonnet 5.5 costs exactly what Sonnet 5 did per token, and it gets more done with each one. At list rates it costs half as much as Opus 5.5, and it trails Opus 5.5 by only a few points on most of Anthropic's launch benchmarks.

The list price is the easy part. What you pay each month depends on four settings most pricing pages skip: batching, caching, where your requests run, and how hard you let the model think.

Get those right and Sonnet 5.5 pricing works out to cents per task. Get them wrong and the same work can cost up to 18 times more. If you are still deciding between models, the Sonnet 5.5 benchmarks cover how its quality compares with Opus 5.5.

Sonnet 5.5 pricing is $2 per million input tokens and $10 per million output

Sonnet 5.5 charges $2 for every million tokens you send and $10 for every million it writes back. A token is a small chunk of text, usually a word or part of one.

Output costs five times as much as input, so long answers drive the bill more than long prompts. The Claude pricing docs list the full rate card.

Table 1 - Claude Sonnet 5.5 API pricing from Anthropic, pricing as of September 2026.

Sonnet 5.5 costs the same as Sonnet 5 and half as much as Opus 5.5

Anthropic kept Sonnet 5.5 at the Sonnet 5 price while raising its scores sharply. Opus 5.5 costs exactly double on input, output, and cache writes.

Table 2 - Anthropic model pricing compared, pricing as of September 2026.

The same price does not mean the same bill. Anthropic's launch post says Sonnet 5.5 costs up to 30% less per task than Sonnet 5 for most work in its testing, since it needs fewer tokens to finish a job.

Opus 5.5 cut its token prices at launch, which narrows the gap. Our Opus 5.5 pricing guide covers the current rates.

Five pricing rules change what you actually pay

Five settings on Anthropic's rate card can cut your bill in half or add 10% to it. None of them change the quality of the answers.

1. Batch API cuts every rate in half

Requests sent through the Batch API cost $1 per million input tokens and $5 per million output tokens. The trade-off is time: batch jobs run asynchronously, so results are not instant.

Batching suits work nobody waits on. Nightly report runs, bulk tagging of support tickets, and weekly summaries of CRM notes are all good fits.

2. Prompt caching pays for itself after one reuse

Caching lets the model reuse text it has already read, such as your instructions or a product catalog. Cache hits cost $0.20 per million tokens, 90% less than the $2 standard input rate.

Writing to the cache costs extra, so the math matters. A 5-minute cache write costs $2.50 per million tokens. Send the same context twice without caching and you pay $4. With caching you pay $2.50 plus $0.20, or $2.70, so the 5-minute cache saves money from the first reuse.

The 1-hour cache costs $4 to write. It pays off after two cache-hit follow-up requests ($4.40 against $6 uncached) and suits context you return to across a longer session.

3. The full 1M context window has no surcharge

Sonnet 5.5 charges the same per-token rate for a 900k-token request as for a 9k-token one, according to Anthropic's docs. There is no long-context premium to plan around.

Long prompts still cost more in total, since you pay for every token. A 200,000-token contract review costs about $0.40 in input alone.

4. US-only inference adds 10%

On the Claude API, setting the inference_geo parameter to "us" keeps processing inside the United States and applies a 1.1x multiplier to every token price. Input rises to $2.20 and output to $11 per million tokens.

The default global routing uses standard pricing. Choose US-only inference when a customer contract or regulation requires it, not by default.

5. Regional cloud endpoints add 10%

Anthropic's launch post says Sonnet 5.5 is available on Amazon Web Services, Google Cloud, and Microsoft Azure, and regional pricing depends on the route. On Amazon Bedrock and Google Cloud, regional and multi-region endpoints carry a 10% premium over global endpoints. Azure deployments can use a US Data Zone option that Anthropic's docs treat as equivalent to US-only inference.

These modifiers combine. On the Claude API, a batch request with US-only inference costs $1.10 per million input tokens and $5.50 per million output tokens.

Everyday tasks cost from about 1 cent to under 50 cents

Most business tasks cost well under a dollar at Sonnet 5.5's list rates. The figures below are calculated from Anthropic's published prices using typical token counts, not measured bills.

Table 3 - Estimated Sonnet 5.5 and Opus 5.5 cost per task at standard list rates, pricing as of September 2026. Token counts are illustrative assumptions. Estimates cover text input and output only, and exclude caching, tool fees, data-residency multipliers, and taxes.

Two settings move these numbers most. Because Table 3 covers input and output only, running the same jobs through the Batch API halves each Sonnet 5.5 estimate. Caching works differently on the first request than on later ones. If 2,000 tokens of the support reply are reusable instructions, the first request costs about $0.012 because it includes the cache write. Each later cache-hit request costs about $0.0074, compared with $0.011 uncached.

The other variable is reasoning. Higher effort settings produce more output tokens, and output is the expensive side of the rate card.

Effort level moves cost per task more than token price does

Sonnet 5.5 has five effort settings, and each step up buys a better answer for a steeper price. On Artificial Analysis's release page, estimated cost per Intelligence Index task rises from $0.41 at Low to $7.60 at Max, about 18 times higher.

The gap comes from output tokens. At Max, Sonnet 5.5 used more output tokens per task than any model Artificial Analysis had measured, which is why its Max-effort cost runs about 50% above Sonnet 5.

At Medium, the default in the Claude apps and Claude Code, Sonnet 5.5 already outscores Sonnet 5's best on that index for about $0.59 per task. The Claude Platform API defaults to High. For most business apps, Medium or High is where Sonnet 5.5 earns its price.

You can pay for Sonnet 5.5 through the API, a Claude plan, or a cloud platform

Sonnet 5.5 is sold three ways, and each suits a different kind of user. The API bills per token, Claude plans bill a monthly fee with usage limits (paid plans can add usage credits beyond them), and cloud platforms bill through your existing cloud account.

Table 4 - Ways to pay for Claude Sonnet 5.5, pricing as of September 2026.

According to Claude's plans page, usage limits on every plan reset on a rolling five-hour window, and paid plans add weekly limits. The plans page lists Sonnet models by family rather than by version number. Plan usage is pooled across models and features, so it can't be turned into a fixed cost per task the way API pricing can.

Opus 5.5 is worth double the price only for judgment-heavy work

Sonnet 5.5 is the better value for most well-scoped tasks. On Anthropic's launch benchmarks it scores 55.5% against Opus 5.5's 57.8% on CursorBench and 1844 against 1846 on GDPval-AA, at half the token price. Opus 5.5 earns its premium on complex, open-ended work that needs sustained judgment, which Anthropic says remains its clear strength.

Opus 5.5 also answers more factual questions correctly. Artificial Analysis measured 66% accuracy for Opus 5.5 on AA-Omniscience against 54% for Sonnet 5.5. For a research assistant or a domain lookup tool, the extra cost can pay for itself in fewer wrong answers.

Price narrows at the top of the effort scale. Anthropic says Sonnet 5.5 at its highest settings can cost about the same per task as Opus 5.5. If a job needs Max effort on Sonnet 5.5, compare it with Opus 5.5 first.

Also read our Opus 5.5 vs Sonnet 5 comparison for which workloads each tier actually suits.

Build with Sonnet 5.5 on Emergent without managing API keys

Claude Sonnet 5.5 pricing rewards teams that set it up well. The token rates are low and fixed, so batching repeat jobs, caching shared instructions, and running Medium or High effort decide most of the bill. Save Opus 5.5 for work that needs its judgment.

Sonnet 5.5 is available on Emergent through the Universal LLM Key. Apps you build can call it without a separate Anthropic account or API key, and usage is billed through Emergent Credits. That makes it a practical fit for a client portal that drafts support replies or an internal tool that summarizes documents at volume. If your work leans toward harder judgment calls, consider Opus 5.5 and where its extra cost pays off.

Start Building on Emergent.