aiapicostcalcUnit economicsPricesBrandsCompareBenchmarksSkills & PluginsFind my modelGuidesChanges

Claude Sonnet API pricing: Sonnet 5, 5.5 and 4.6

Sonnet 5 and Sonnet 5.5 cost $2 per 1M input and $10 per 1M output tokens; Sonnet 4.6 costs $3/$15. A 4K-in / 600-out support answer costs $0.0140 on Sonnet 5.5, $140 for 10,000 answers.

Last verified 2026-11-18 · Sources: official provider pricing pages, copied into src/prices.json (checked 2026-11-18) · Report an error

By Shangyi Duan · Reviewed against Anthropic pricing page

Source URL: https://docs.anthropic.com/en/docs/about-claude/pricing · Verified: · Rate row: src/prices.json

1
2
Presets

Preset token counts are starting assumptions; replace them with your logged averages.

Cost per call$0.0140
Per 1,000 calls$14.00

$2 in · $10 out per 1M

Nearest-priced models, this workload

Showing the 9 models closest in price; change any input to rank every model.

ModelPer callPer 1K calls
Grok 4.7$0.0116$11.60
GPT-4.1 (deprecated)$0.0128$12.80
GPT-6.1 Sol$0.0140$14.00
GPT-6 Sol$0.0140$14.00
Claude Sonnet 5.5$0.0140$14.00
Claude Sonnet 5$0.0140$14.00
Gemini 4 Carbon$0.0140$14.00
Gemini 4 Argon$0.0140$14.00
GPT-5.6 Terra$0.0152$15.20

o4-mini is shut down; its last published list rate is kept for comparing old invoices, not as a model you can call today — the model page names its replacement.

What Claude Sonnet costs on the Anthropic API

Sonnet 5 and Sonnet 5.5 both list at $2 per 1M input tokens and $10 per 1M output tokens. Sonnet 4.6, the previous generation, is $3/$15. The list prices of 5 and 5.5 are identical; the cache-read rate is not ($0.1 on 5.5 against $0.2 on 5).

ModelInput /1MCached read /1MOutput /1MChecked
Claude Sonnet 5.5$2$0.1$102026-11-15
Claude Sonnet 5$2$0.2$102026-11-15
Claude Sonnet 4.6$3$0.3$152026-11-15

Rates copied from the Anthropic pricing page, checked on the dates in the table.

Sonnet 5 vs 5.5 vs 4.6

Sonnet 5 and 5.5 are the same price per token; 5.5 halves the cache-read rate, which matters for RAG and agent prompts that reuse a long prefix. Sonnet 4.6 costs 50% more on both input and output than either 5 model, so keep it only where an evaluation set or a pinned prompt depends on it. There is no published reason to pay the extra 50% for a new workload.

What a Claude Sonnet workload costs

Three common production shapes, priced at standard list rates. The calculator above lets you change the token mix; the numbers below are the arithmetic behind it.

WorkloadSonnet 5.5 per callSonnet 5 per callCalls per monthSonnet 5.5 monthlySonnet 5 monthly
Support answer (4K in / 600 out)$0.0140$0.014010,000$140$140
RAG answer, 80% of the prompt cached$0.0196$0.02121,000$19.60$21.20
Batch summarisation (3K in / 400 out)$0.0100$0.01005,000$50.00$50.00

The batch row uses the standard rate; the Batch API discount is 50% off where Anthropic publishes it, which would halve that line again.

Where Sonnet fits

Sonnet is the default production tier: chat, RAG, extraction and batch jobs where Opus quality is not what limits the answer. Move up to Opus 5.5 when the task needs several reasoning steps or the cost of a wrong answer is high, and down to Haiku 5.5 for classification, routing and other high-volume, short-output work.

Sonnet 5.5 at production volume

Sonnet 5.5 lists $2 per 1M input and $10 per 1M output tokens. It is the tier most chat, retrieval and extraction products settle on, so the useful question is what a month of that traffic costs. The three shapes below use the same token counts throughout the site.

A 10,000-token prompt with a 1,000-token answer

One call costs $0.0300, 1,000 calls cost $30.00. Output is 33% of the bill despite being a tenth of the tokens. Claude Sonnet 5 lists the same two rates, so that request costs $0.0300 on it: the list prices of 5 and 5.5 are identical.

A retrieval answer: 5,000 context tokens in, 2,000 out

This costs $0.0300 per answer and $30.00 per 1,000 answers. On a RAG product the cheapest saving is usually the prompt, not the model: cache reads on Sonnet 5.5 bill at $0.1 per 1M against $2 standard, and Sonnet 5 bills $0.2 for the same read. That difference is why 5.5 is the version to deploy when a fixed prefix is re-sent, and the prompt caching guide shows the break-even on a 20K prefix.

A 40-turn agent session

With a 60,000-token prefix written to cache on turn one, read cached on the next 39 turns, and 3,000 fresh input plus 2,000 output tokens per turn, one session costs $1.42 and 1,000 sessions cost $1,418. Output is 1410437% of it. The same session on Sonnet 5 costs $1.65: the two models share a rate card, but the cached-read price ($0.1 against $0.2) takes 14% off a cache-heavy run. The 5-minute cache write is $2.5 per 1M and the 1-hour write $4.

Cost per user per month

At 200 calls a month on the 10,000/1,000 mix one user costs $6.00 in model time; at an 80% gross margin the minimum price is $30.00 per user per month. At 500 calls it is $15.00 of cost and $75.00 of price. The AI SaaS pricing guide covers the tail users that an average-based price always loses money on.

Sonnet price checks since 4 November 2026

The official Anthropic pricing page has been fetched and diffed on 101 completed checks since our monitoring began on 2026-11-04, and none of those checks flagged a change to the list prices above. The row used here was last confirmed against that page on 2026-11-15. Failed fetches are skipped rather than counted, so the number is checks, not days.

What moves a Sonnet bill without a list-price change: the Batch API discount of 50% on asynchronous jobs, and cache economics. The batch guide prices the first; the token estimation guide is the place to start if your request logs do not yet separate input from output tokens, because a bill model without that split is guesswork.

Sonnet 5.5 against Opus 5.5, Haiku 5.5 and the other 2/10 models

Prices in this band cluster: Sonnet 5.5 lists $2/$10, Gemini 4 Argon lists $2/$10 on the same two numbers, and GPT-5.4 lists $2.5/$15. The 10,000/1,000 call costs $0.0300 on Gemini 4 Argon and $0.0400 on GPT-5.4 at those rates. One tier up, Claude Opus 5.5 lists $4/$20, double on both sides: $0.0600 for the same call, $2.84 for the 40-turn session against $1.42. One tier down, Claude Haiku 5.5 lists $0.1/$0.5: $0.0015 for the call and $0.0826 for the session.

Move up to Opus when a wrong answer costs more than the rate gap, and down to Haiku for routing, classification and short extraction where the answer is a label rather than prose. Head-to-head pages: GPT-5.4 vs Claude Sonnet 5.5, Claude Sonnet 4.6 vs Claude Sonnet 5.5 and Claude Sonnet 5.5 vs Gemini 4 Argon. For subscription pricing from measured usage, the AI SaaS pricing guide pairs with the unit economics calculator; Claude cost calculator runs your own token mix.

Related Claude prices

Questions

How much is Claude Sonnet per 1M tokens?

Sonnet 5 and Sonnet 5.5 are $2 input and $10 output per 1M tokens. Sonnet 4.6 is $3 input and $15 output. Cache reads are $0.1 on 5.5 and $0.2 on 5.

What is the difference between claude sonnet pricing and claude sonnet api cost?

The pricing is the rate card, the cost is the rate times your tokens. Same list price does not mean the same bill: a cached RAG prompt and a fresh chat prompt at the same token count cost different money.

Should I use Sonnet 5 or Sonnet 5.5?

Use 5.5 for anything new. The list price is identical and the cache-read rate is half, so the only way 5 comes out cheaper is a prompt mix with no reusable prefix.

Is Sonnet 4.6 more expensive than Sonnet 5?

Yes, 50% more on both rates ($3/$15 against $2/$10). It stays on the site as its own page because it still carries search demand, not because it is the cheaper choice.

Do cached and batch rates change the Sonnet bill?

Both can. A cache hit is billed at the cached-read rate instead of the input rate, and the Batch API is half price where Anthropic publishes it. Neither is switched on by default in the calculator; enter a cached % or check the batch column on the model page.

Next step → Estimate cost per user with Claude Sonnet

Price data last checked 2026-11-18 against official sources · how we verify

Price-change alerts

One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.