Claude Sonnet 5.5 pricing
Claude Sonnet 5.5 API cost from Anthropic — $2.00 per 1M input tokens and $10.00 per 1M output tokens. Standard rates across 1M context; 5-minute and 1-hour cache writes billed separately. Estimate your real monthly bill below, then compare it against every other model.
Balanced 1M context · all prices USD per 1M tokens, this price verified 2026-10-09.
Cache writes: $2.50 / 1M tokens. Batch: 50% of Standard.
Sonnet 5.5 rates, including cache writes
Anthropic's Sonnet 5.5 overview lists the API model as claude-sonnet-5-5, with a 1M-token context window and up to 128K output tokens. The prices below were checked on October 9, 2026 and apply to the first-party Claude API with global routing.
| Billing category | USD per million tokens |
|---|---|
| Fresh input | $2 |
| Billable output | $10 |
| Cache reads | $0.10 |
| 5-minute cache writes | $2.50 |
| 1-hour cache writes | $4 |
| Batch input / output | $1 / $5 |
Sonnet keeps its Standard rate across the full context window. A 900K-input request costs more than a 9K-input request because it contains more tokens, but the price per token stays the same. Do not apply Haiku 5.5's 100K price boundary to Sonnet.
A $60 workload you can reproduce
Take 10,000 independent requests, each with 2,000 fresh input tokens and 200 billable output tokens. One request costs (2,000 × $2 + 200 × $10) ÷ 1,000,000 = $0.006. The batch of 10,000 costs $60 at Standard rates, or $30 if eligible for the Batch API discount.
Open Sonnet with those inputs. The calculator applies Standard input, output and optional cache-read rates. Cache writes, Batch, tools, residency and taxes remain outside its estimate. Enter the total billable output reported by the API, including thinking, rather than just the visible answer.
Cache reuse: pay for creation before counting savings
Suppose a reusable document contains 10,000 tokens. The first request writes that document to a 5-minute cache, adds 1,000 fresh tokens, and produces 500 output tokens. Its cost is $0.025 for the write + $0.002 fresh input + $0.005 output = $0.032.
A later request reading all 10,000 cached tokens, adding the same 1,000 fresh tokens and producing the same output costs $0.001 + $0.002 + $0.005 = $0.008. Ten such requests, one write followed by nine hits, total $0.104. Ten uncached requests with all 11,000 input tokens would cost $0.27.
Those are planned cache hits within the cache lifetime, not a promise that every call will hit. With a 1-hour cache, the first request instead costs $0.047 and the ten-request total becomes $0.119. A cache refresh or changed prefix can add another write. Bill each input token once in its actual category; adding the ordinary input price on top of a cache-write price would double-count it.
Batch and regional processing
The official price list gives a 50% Batch discount and says caching multipliers stack with it. A successful cache hit is still subject to the cache's lifetime and eligibility rules. Asynchronous work can be cheaper, but a queue is a poor fit when a person needs an immediate reply.
US-only inference adds 10% to token categories. The uncached $60 example becomes $66 if that setting applies. Partner-operated Amazon Bedrock and Google Cloud offerings can have independent regional rates. Identify the platform and billing mode before comparing an invoice with this page.
Budget for a completed workflow
An agent often makes several model calls to solve one user request. Include the conversation sent again on later turns, tool definitions and tool results. If a ten-turn workflow repeats a long document without a cache, the document can be billed many times.
Adaptive thinking is supported on Sonnet 5.5. A short final response does not imply a small output bill when the model uses thinking. Paid server tools can also add their own charges. Compare the API's usage breakdown with the configured mode before attributing a higher bill to the advertised base rate.
For a monthly forecast, group requests by their input and output distributions, then add retries and cache setup. Multiplying one unusually short successful request by monthly traffic will understate a workload with long prompts or frequent retries.
API prices, subscriptions and resellers
A Claude consumer or team subscription is billed separately from metered API usage. Included allowances or API credits depend on the plan; they do not change Sonnet's token list price. Check the plan's current terms before treating any credit as recurring budget coverage.
This page lists Anthropic's API rates. A reseller may add a markup, offer a temporary subsidy, bill in a different currency or use different routing. Match the exact model ID, cache policy and service tier before deciding that two advertised prices represent the same purchase.
When to compare Sonnet with Haiku or Opus
For focused extraction and classification, compare Haiku 5.5 on examples with known answers. For longer coding or knowledge tasks, compare Opus 5.5 on your acceptance criteria. A lower unit price alone does not prove a lower cost per accepted result.
Use the Anthropic price table to check adjacent models, and keep the original usage totals when changing versions. Token counts and the number of calls can change even when you reuse the same application prompt.
Claude Sonnet 5.5 capabilities & limits
Representative specs for this tier — confirm exact limits in the Anthropic docs.
API acting up? Check live Anthropic status → · all AI API status
Call Claude Sonnet 5.5 from your code
The exact model string and endpoint to use. Uses the Anthropic API format.
claude-sonnet-5-5https://api.anthropic.com/v1ANTHROPIC_API_KEYcurl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-sonnet-5-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Hello"}]
}'Model strings follow each provider's naming convention; confirm the current id in the Anthropic docs before relying on it.
What a Claude Sonnet 5.5 API call actually costs
Blended input + output cost at Claude Sonnet 5.5's list rates — useful yardsticks before you commit to volume.
Claude Sonnet 5.5 cost calculator
Enter your monthly volume to estimate the Claude Sonnet 5.5 bill. Runs entirely in your browser.
Peak/off-peak and long-context pricing are applied automatically where published. Cache-write, Batch/Flex and Fast-mode adjustments are not included. For the multi-model breakdown with the same exclusions, use the cost calculator.
Anthropic models — Claude Sonnet 5.5 in context
How Claude Sonnet 5.5 prices against its siblings. Cheapest reference cost first.
| Model | Input /1M | Output /1M | Cost / call* | Context | Type |
|---|---|---|---|---|---|
| Claude Haiku 5.5 Prompts up to 100K; above 100K, all token rates are 5× higher | $0.100 | $0.500 | $0.0006 | 1M | Fast / cheap |
| Claude Haiku 4.5 Fastest, near-frontier | $1.00 | $5.00 | $0.0060 | 200K | Fast / cheap |
| Claude Sonnet 5.5 Standard rates across 1M context; 5-minute and 1-hour cache writes billed separately | $2.00 | $10.00 | $0.012 | 1M | Balanced |
| Claude Sonnet 4.6 Best speed/intelligence; caching cuts input ~90% | $3.00 | $15.00 | $0.018 | 1M | Balanced |
| Claude Opus 5.5 Standard rates across 1M context; Fast costs 2× and cannot combine with Batch | $4.00 | $20.00 | $0.024 | 1M | Flagship |
| Claude Opus 4.8 Top Opus reasoning/agentic | $5.00 | $25.00 | $0.030 | 1M | Flagship |
| Claude Fable 5.1 Standard rates across 1M context; cache writes billed separately | $10.00 | $50.00 | $0.060 | 1M | Flagship |
| Claude Fable 5 Most capable widely released | $10.00 | $50.00 | $0.060 | 1M | Flagship |
*Reference cost of one call with 1,000 input + 1,000 output tokens — a neutral yardstick. Use the calculator for your real usage. Cheapest row highlighted.
How Claude Sonnet 5.5 pricing works
Claude Sonnet 5.5 is billed per token: a lower input price of $2.00 per million tokens for everything you send, and a higher output price of $10.00 per million for everything the model generates. To turn that into a real Claude Sonnet 5.5 bill, multiply by your monthly request volume and average prompt/response length — the calculator above does exactly that.
Prompt caching is the biggest lever on your Claude Sonnet 5.5 cost: cached input tokens bill at just $0.100 per million — about 95% cheaper. If you reuse a long system prompt or context across calls, caching can dramatically cut the input portion of your bill.
List prices move fast — always confirm the current numbers on the official Anthropic pricing page before relying on them for budgeting.
Claude Sonnet 5.5 pricing FAQ
How much does the Claude Sonnet 5.5 API cost?
Claude Sonnet 5.5 costs $2.00 per 1M input tokens and $10.00 per 1M output tokens, with cached input at $0.100 per 1M (about 95% off). A typical 3K-input / 500-output call at the displayed rate works out to about $0.011.
What is Claude Sonnet 5.5's context window?
Claude Sonnet 5.5 has a 1M-token context window. Standard uncached input is $2.00 per 1M tokens.
Does Claude Sonnet 5.5 support prompt caching?
Yes. At Standard rates, Claude Sonnet 5.5 bills cache-read tokens at $0.100 per 1M instead of $2.00 uncached — roughly 95% cheaper for the repeated part of your prompts.
Is there a cheaper alternative to Claude Sonnet 5.5?
In the same balanced tier, Command R (Cohere) is cheaper at $0.150 in / $0.600 out. For the 1K-input / 1K-output reference workload, the lowest-cost model tracked in this catalog is Gemini 1.5 Flash-8B.