Claude Sonnet 5.5 API pricing and cost calculator

By TechCompare · Updated · How we check

A Sonnet request with 100K input and 5K output costs $0.25 uncached or $0.155 with half the input cached. Prompt growth raises token volume without adding a long-context price tier.

Calculator

Cost Comparison

Based on 100,000 input tokens (50% cached), 5,000 output tokens, and 100 requests.Prices come live from OpenRouter at each maker's own price where it's listed. Temporary deals are left out.

Loading model prices

Claude Sonnet 5.5 charges $2 per million input tokens and $10 per million output tokens at standard rates. Its $0.10 cache-read price reduces repeated-context spending, while standard token rates stay flat within its 1M context window.

Those properties are useful when budgeting document processing or ongoing conversations. The bill still depends on how much new input and output each turn adds, so start with actual usage rather than a model's context capacity.

Anthropic: Claude Sonnet 5.5

Anthropic's own price

Input
$2
Output
$10
Cached
$0.10

per 1M tokens

Batch available

Price checked 2026-11-19

How this is calculated

At 100,000 input and 5,000 output tokens, Sonnet costs $0.20 for uncached input plus $0.05 for output, totaling $0.25. The starting calculator example reads half the input from cache, giving $0.10 uncached input, $0.005 cached input, and $0.05 output. That's $0.155 per request or $15.50 for 100 monthly requests.

A 300,000-input, 5,000-output prompt costs $0.65 with no cache hits or $0.365 with half the input cached. A longer prompt doesn't introduce a separate token-rate tier within the advertised window. These examples exclude cache writes and tool charges. The embedded estimate uses OpenRouter's live provider rates, so check its routing assumptions alongside Anthropic's standard price list.

Verdict

Sonnet is worth testing when its results meet your quality targets and a predictable token rate helps you budget long prompts. Keep output concise when the task allows it, and compare retrieval with sending the full document on every turn.

Prompt caching is valuable for reusable prefixes, but count the writes and misses too. If another model completes a task in fewer calls, its total session cost can matter more than a small difference in input rates.

More API Standalones scenarios

Frequently asked questions

What is the cost of 10,000 Sonnet 5.5 requests?
For the 100,000-input, 5,000-output example, 10,000 requests cost $2,500 with no cache hits or $1,550 with half the input cached. Those are token estimates, excluding writes, tool charges, and retries. For a real budget, group requests by their actual prompt lengths and cache usage instead of assuming every request looks like the same example.
How much does Sonnet 5.5 prompt caching cost?
Reads cost $0.10 per million tokens, a 95% reduction from the uncached input rate. Five-minute writes cost $2.50 per million tokens and one-hour writes cost $4, twice the input rate. Cache reuse needs to recover those creation costs. Estimate the number of hits before expiration and avoid treating an input prefix that changes each turn as a guaranteed read hit.
How does Sonnet 5.5's price compare with GPT-6.1 Sol?
Their base input and output rates match. Their cache-read prices match, while Sonnet avoids Sol's prompt-length surcharge within its own context window. Run your own prompt lengths, cache usage, and output volume through both before deciding. The two providers can also bill different token counts for the same text, so measure usage on your own requests.
Does Sonnet's 1M context window mean I should use it all?
No. Capacity is a limit, not a spending target. Sending more material raises input volume even when the per-token rate stays flat. Keep relevant document sections and tool results, remove needless repetition, and test whether the shorter prompt preserves task quality. A targeted prompt may cost less and be easier to evaluate than a large collection sent on every turn.
What does Sonnet 5.5 cost in batch mode?
Anthropic's Batch API cuts applicable token prices by 50%. The uncached example drops from $0.25 to $0.125 per request, and the half-cached token example drops from $0.155 to $0.0775 before write costs and tools. That's an asynchronous provider workflow. The calculator models those rates when batch mode is enabled, rather than claiming OpenRouter's live requests have the same discount.