Anthropic API pricing 2026: 14 Claude models

According to APITariff, Anthropic API pricing on 2026-10-10 runs, across its 14 current models, from $0.10 input and $0.50 output per 1M tokens on Claude Haiku 5.5, for prompts up to 100,000 tokens, to $10 and $50 on Claude Fable 5.1. Over 1,000,000 calls of 1,000 input and 300 output tokens the bill is $250 on Haiku 5.5, $2,500 on Claude Haiku 4.5, $5,000 on Claude Sonnet 5.5, $10,000 on Claude Opus 5.5 and $25,000 on Fable 5.1. Haiku 5.5 bills a request whose prompt is over 100,000 tokens at $0.50 and $2.50, 5 times its base rate.

kie.ai lists Haiku 5.5 at $0.06 input and $0.30 output, so the same calls bill $150 there against $250 direct. Anthropic’s rows match OpenRouter’s on all 16 Claude slugs it lists, the Haiku 5.5 slug carrying only its up-to-100,000-token rate. None of the 18 Claude slugs we saw there changed price across up to 29 fetches between 2026-09-02 and 2026-10-10, but 2 aliases moved.

Last updated 2026-10-10 · 14 current models · sources: Anthropic pricing page, APITariff fetch of OpenRouter and kie.ai, 2026-10-10.

Anthropic API pricing per 1M tokens

Of the 21 rows on Anthropic’s pricing page, 15 belong to the 14 current models open to any account, because Haiku 5.5 has one row per prompt-length tier. The other 6 are 4 retired models (Opus 4.1, Opus 4, Sonnet 4, Haiku 3.5) and 2 Mythos models with limited availability. Opus 4.1 and Opus 4 list the highest rate on the page, $15 input and $75 output.

Model	Input $/1M	Output $/1M	Cache hit $/1M	Batch in / out $/1M
Claude Fable 5.1	10	50	0.25	5 / 25
Claude Fable 5	10	50	1	5 / 25
Claude Opus 5.5	4	20	0.20	2 / 10
Claude Opus 5	5	25	0.50	2.50 / 12.50
Claude Opus 4.8, 4.7, 4.6, 4.5	5	25	0.50	2.50 / 12.50
Claude Sonnet 5.5	2	10	0.10	1 / 5
Claude Sonnet 5	2	10	0.20	1 / 5
Claude Sonnet 4.6, 4.5	3	15	0.30	1.50 / 7.50
Claude Haiku 4.5	1	5	0.10	0.50 / 2.50
Claude Haiku 5.5, prompts over 100K tokens	0.50	2.50	0.05	0.25 / 1.25
Claude Haiku 5.5, prompts up to 100K tokens	0.10	0.50	0.01	0.05 / 0.25

14 of 14 current models shown, 11 rows. Source: Anthropic pricing page, read 2026-10-10.

Three details change a bill. A cache hit costs 0.1x the input price on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.1. Batch processing is 50% off both input and output.

The third is prompt length on Haiku 5.5: a request whose prompt is over 100,000 tokens pays 5 times the base rate on every field in the table. Anthropic says Claude 4.6 and later models, except Haiku 5.5, bill a 900,000-token request at the same per-token rate as a 9,000-token one.

How much does the Claude API cost per month?

At 100,000 calls a month of 1,000 input and 300 output tokens the bill is $25 on Haiku 5.5 and $2,500 on Fable 5.1. Every current model lists its base output rate at 5 times its base input rate, so both scale together: Sonnet 5.5 costs half of Opus 5.5, and Fable 5.1 costs 5 times Sonnet 5.5.

Model	10k calls	100k calls	1M calls	1M calls, batch
Haiku 5.5	$2.50	$25	$250	$125
Haiku 4.5	$25	$250	$2,500	$1,250
Sonnet 5.5	$50	$500	$5,000	$2,500
Sonnet 4.6	$75	$750	$7,500	$3,750
Opus 5.5	$100	$1,000	$10,000	$5,000
Opus 5	$125	$1,250	$12,500	$6,250
Fable 5.1	$250	$2,500	$25,000	$12,500

Computed from the first table. Workload: 1,000 input tokens and 300 output tokens per call, no caching, so Haiku 5.5 bills at its up-to-100,000-token rate. A call with a 150,000-token prompt and 1,000 output tokens bills $0.0775 on Haiku 5.5 and $0.155 on Haiku 4.5.

The tables use equal token counts, not equal text, and Anthropic says Claude 4.7 and later models produce about 30% more tokens for the same text. If your prompts are longer than 1,000 tokens, the token calculator takes your own token counts.

Fast mode is a research preview on Opus 5.5, Opus 5 and Opus 4.8. On Opus 5.5 it lists $8 input and $40 output per 1M tokens, which bills $20,000 for 1,000,000 calls. Anthropic’s page says fast mode runs on the first-party API only and not with the Batch API.

What does prompt caching save on Claude?

On a Claude API call with 90,000 cached input tokens, 10,000 uncached input tokens and 1,000 output tokens, caching cuts the bill by 77.1% on Haiku 5.5, 81.4% on Sonnet 5.5 and Opus 5.5 and 83.6% on Fable 5.1, computed from Anthropic’s rates. The prefix is assumed already cached, so the one-time write cost is excluded. Anthropic’s pricing page states that a 5-minute cache write at 1.25x pays back after one read, and a 1-hour write at 2x after two reads.

Model	With cache hits	No cache	Saving	10,000 calls, cached
Haiku 5.5	$0.0024	$0.0105	77.1%	$24
Haiku 4.5	$0.024	$0.105	77.1%	$240
Sonnet 5.5	$0.039	$0.21	81.4%	$390
Opus 5.5	$0.078	$0.42	81.4%	$780
Opus 5	$0.12	$0.525	77.1%	$1,200
Fable 5.1	$0.1725	$1.05	83.6%	$1,725

Computed from the first table. “No cache” bills all 100,000 input tokens at the base input rate. The Haiku 5.5 call sits exactly at the 100,000-token line and uses the base tier; Anthropic’s page says a request over 100,000 prompt tokens pays the higher prices, so 100,001 tokens would bill the whole call at the higher tier.

Sonnet 5.5’s cache hit is now $0.10, after Anthropic halved it with the Haiku 5.5 launch (page dated 2026-10-07), so it sits at half of Opus 5.5’s $0.20. Opus 5.5 therefore costs 2.0x Sonnet 5.5 on both the uncached and the cached call, and Fable 5.1 costs 4.42x Sonnet 5.5 on the cached call, down from 5.0x uncached.

Which Claude model costs least?

Claude Haiku 5.5 costs least among the 14 current models for prompts up to 100,000 tokens: $0.10 input and $0.50 output per 1M tokens on 2026-10-10, which bills $250 per 1,000,000 calls. The rule is the lowest output price per 1M tokens, and ranking by input price gives the same order because every current model lists output at 5 times input.

Ranked that way, Haiku 5.5 is lowest at $0.50 up to 100,000 prompt tokens and $2.50 above, then Haiku 4.5 at $5, Sonnet 5.5 and Sonnet 5 at $10, Sonnet 4.6 and 4.5 at $15, and Opus 5.5 at $20. APITariff did not bench Claude models, so this ranks price and says nothing about quality or speed; test the model on your own task before you commit volume.

Haiku 5.5 bills 90% less than Haiku 4.5 up to 100,000 prompt tokens and 50% less above, the same two figures Anthropic prints on its launch page. The Haiku 5.5 vs GPT-6 Luna comparison prices that 100,000-token line against OpenAI’s small model.

Token counts shift these comparisons. Anthropic’s pricing page says Claude 4.7 and later models use a tokenizer that produces approximately 30% more tokens for the same text, and that Sonnet 4.6 and earlier use the previous one.

Sonnet 5.5 lists input at $2 against $3 on Sonnet 4.6, a 33.3% cut. If the text turns into 30% more tokens, the same input costs $2.60 per 1M old-tokenizer tokens, 13.3% below $3, and against Haiku 4.5 at $1 the 2.0x list ratio becomes about 2.6x per unit of text. For Haiku 5.5 against Haiku 4.5 the same adjustment leaves the 1,000 plus 300 token call 87.0% lower, and a prompt of 80,000 Haiku 4.5 tokens becomes about 104,000, over the 100,000-token line.

These figures are estimates built on Anthropic’s “approximately 30%”, and we did not measure token counts on any of your text.

Opus 5.5 against Opus 5 needs no such adjustment. Both use the newer tokenizer, and Opus 5.5 is 20% lower on input and output: $4 and $20 against $5 and $25. The Opus 5.5 pricing note has the field-by-field table of all 9 price fields.

Did Anthropic change Claude API prices?

Only two aliases moved, and no model’s list price did. APITariff’s price-change log holds 2 Claude price entries, both on the ~anthropic/claude-opus-latest alias: input from $5 to $4 and output from $25 to $20 on 2026-09-23, both -20.0%, when the alias began pointing at Opus 5.5.

We counted the input and output prices of the 18 anthropic/claude slugs without a :batch suffix across up to 29 fetches from 2026-09-02 to 2026-10-10 (Opus 5.5 appears in 17 of them, Sonnet 5.5 in 11, Haiku 5.5 in 2), and each slug read one price. 16 of the 18 are still listed; anthropic/claude-opus-4 was last read on 2026-09-21 and anthropic/claude-3-haiku on 2026-09-25. APITariff first fetched on 2026-09-02, so it holds no Claude reading from before that date, and 38 days is too short to state a trend.

The Haiku alias moved too, but the log does not show it. ~anthropic/claude-haiku-latest read $1.00 input until 2026-10-07 and $0.10 on 2026-10-08 and again on 2026-10-10, a 90% drop. Our engine treats moves over 80% as probable fetch errors, so it quarantined both readings instead of logging them, and we have not read the alias’s output price here. Two other Claude rows in the log on 2026-10-10 are kie.ai output fields appearing or vanishing, not price moves.

Two Anthropic statements explain the shape of the Opus move. Opus 5.5 launched on 2026-09-22 at 20% below Opus 5, according to Anthropic’s launch page, and our fetch first read its slug on 2026-09-23. A footnote on Anthropic’s page says Sonnet 5’s $2 and $10 introductory price through 2026-08-31 “is now the standard price”, and that the scheduled increase to $3 and $15 on 2026-09-01 “will not occur”. Sonnet 5.5 first appears in our fetch on 2026-09-29 at the same $2 and $10.

Anthropic’s launch page for Haiku 5.5 is dated 2026-10-07, and our fetch first read its slug on 2026-10-08. The same page halves Sonnet 5.5’s cache-read price, which our OpenRouter rows cannot show, because they hold input and output prices only.

OpenRouter’s model list equals Anthropic’s page on all 16 Claude slugs and on all 15 :batch slugs, which we compared with Anthropic’s batch table. For Haiku 5.5 that equality covers the up-to-100,000-token rate only, because OpenRouter lists one rate per slug.

Is kie.ai cheaper than Anthropic’s own API?

The kie.ai button on this page carries our affiliate code, and it changed none of the figures below. For the models both list, yes, on list price. kie.ai lists Haiku 5.5 at $0.06 input and $0.30 output per 1M tokens, which bills $150 per 1,000,000 calls against $250 direct, and Sonnet 5.5 at $0.80 and $4.00, which bills $2,000 against $5,000 direct and $2,500 on Anthropic’s batch rate (APITariff fetch of kie.ai, 2026-10-10). kie.ai prints a 60.0% discount on its Sonnet 5.5, Opus 5.5 and Opus 5 rows, read from its pricing API on 2026-10-07.

Model	kie.ai in / out $/1M	Anthropic in / out $/1M	kie.ai bill, 1M calls	Anthropic bill	Anthropic batch bill
Haiku 5.5	0.06 / 0.30	0.10 / 0.50 (to 100K)	$150	$250	$125
Haiku 4.5	0.275 / 1.425	1 / 5	$702.50	$2,500	$1,250
Sonnet 5.5	0.80 / 4.00	2 / 10	$2,000	$5,000	$2,500
Sonnet 5	0.85 / 4.275	2 / 10	$2,132.50	$5,000	$2,500
Opus 5.5	1.60 / 8.00*	4 / 20	$4,000	$10,000	$5,000
Opus 5	2.00 / 10.00	5 / 25	$5,000	$12,500	$6,250
Fable 5	4.00 / 20.00	10 / 50	$10,000	$25,000	$12,500

7 of 13 Claude rows on kie.ai shown. *kie.ai’s Opus 5.5 output row carries the label claude-opus-5. Source: APITariff fetch of kie.ai, kie.ai Haiku 5.5 page and Anthropic pricing page, all read 2026-10-10. Each kie.ai price was the same in all 16 kie.ai fetches since 2026-09-24, or in all fetches since the model first appeared (Opus 5.5 in 11 since 2026-09-29, Sonnet 5.5 in 9 since 2026-10-01, Haiku 5.5 in 1, on 2026-10-10).

On list price kie.ai bills 40.0% to 71.9% below the direct rate on these 7 models, and 14.7% to 43.8% below Anthropic’s batch rate on 6 of them. On Haiku 5.5 it bills $150 against $125 on Anthropic’s batch rate, 20.0% more. Five limits apply, and the first may reverse the order for cached calls.

• Cached calls, not measured: kie.ai’s cache terms were not fetched. If kie.ai bills cached tokens at its input rate, the 90,000-cached-token call costs $0.084 on Sonnet 5.5 and $0.168 on Opus 5.5, against $0.039 and $0.078 direct, which is 53.6% lower on Anthropic for both. On Haiku 4.5 the call costs $0.0289 on kie.ai against $0.024 direct, 17.0% lower on Anthropic.
• Long prompts: kie.ai’s Haiku 5.5 page, read 2026-10-10, lists no price for prompts over 100,000 tokens, so the $150 holds for short prompts as far as we can tell.
• Fable 5.1: kie.ai lists Fable 5 and has no Fable 5.1 row, and no row for the Mythos models, Opus 4.1 or Opus 4.
• Opus 5.5 output: kie.ai’s pricing API, read 2026-10-07, labels that row “claude-opus-5” at usdPrice 8.00 with falPrice 20.00, Anthropic’s Opus 5.5 output rate. We read $8.00 as Opus 5.5 output, 40% of 20.00; check the row before you send traffic there.
• Terms: kie.ai is a reseller. We did not fetch its rate limits, refund terms or data-retention terms, and APITariff sent no Claude request through it.

Is the Claude API free?

Not as a standing free tier. The model pricing and batch tables on Anthropic’s page, read 2026-10-10, list no $0 row, and the page’s FAQ says new users receive “a small amount of free credits to test the API”, without an amount. Anthropic’s launch page for Haiku 5.5, dated 2026-10-07, adds a monthly API credit for Claude Max and Team subscribers, also without an amount.

If you need a model at $0, the free AI API guide counts those that list one on other providers and states their limits.

What this page does not cover

Our engine reads OpenRouter, kie.ai and Replicate, never Anthropic’s own API, so the Anthropic rates above come from its pricing page, read by hand on 2026-10-10. Not covered:

• The 2 Mythos models, limited availability, $10 input and $50 output on Anthropic’s page.
• Haiku 5.5 prompts over 100,000 tokens in the bill tables: they use the up-to-100,000-token rate, and the higher rate is in the first table.
• Amazon Bedrock and Google Cloud prices. Anthropic’s page says regional and multi-region endpoints add a 10% premium over global ones on Claude 4.5 models and later.
• The 1.1x multiplier for US-only inference, tool use and web search charges.
• Claude subscription plans, speed and output quality.

To see where these rows sit among every text model, open the cheapest LLM API ranking. The Claude vs OpenAI comparison runs the same 1,000 plus 300 token workload against 8 models, and the methodology page states how each fetch works. The full kie.ai price list covers every kie.ai model in the feed.

Cite as: According to APITariff (apitariff.com/guides/anthropic-api-pricing/, read 2026-10-10).

A person checked and edited this AI-assisted draft of APITariff’s own price data. The chart at the top is drawn from the tables on this page and is not a generated image.

Change log

• 2026-10-10Refresh after Claude Haiku 5.5: added Haiku 5.5 (two prompt-length rows), Sonnet 5.5 cache hit $0.20 to $0.10, Anthropic's free-credits statement, kie.ai Haiku 5.5, fetch counts to 2026-10-10, the quarantined Haiku alias, new hero chart. All other Anthropic rows re-read and unchanged.
• 2026-10-07First edition. 13 current models read on Anthropic's pricing page, compared with OpenRouter and kie.ai on 2026-10-07.