Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Cost levers - verified

How to cut your Anthropic API bill

Prompt caching: explicit only. Cache lifetime: 5 min, refreshed on each hit, or 1 hour. A cache hit typically saves 90% on input across 15 current models. Batch: 50% off, 24h (most under 1 hour). Batch and caching stack: yes. Service tiers: Fast mode (2x), Priority Tier (commitment only, no new sales).

Read from platform.claude.com. Compare every provider's cost levers

Caching
Explicit only
Cache hit saves
90%
Batch
50% off
Service tiers
2

Prompt caching

Mode

Explicit: nothing is cached until you mark it

Opt-in via cache_control: either one top-level cache_control (automatic breakpoint placement) or explicit block-level breakpoints (up to 4). No caching without cache_control.

Cache lifetime

5 minutes by default, refreshed for no additional cost on each hit; 1-hour option at 2x base input

Minimum prompt

512 tokens: Fable 5.1, Mythos 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Fable 5, Mythos 5, Haiku 5.5. 1,024: Opus 4.8, Sonnet 5, Sonnet 4.6, Sonnet 4.5, Opus 4.1, Opus 4, Sonnet 4. 2,048: Mythos Preview, Opus 4.7, Haiku 3.5. 4,096: Opus 4.6, Opus 4.5, Haiku 4.5.

Cache write cost

5-minute writes 1.25x base input; 1-hour writes 2x base input

Cache read cost

0.1x base input (0.025x on Fable 5.1 and Mythos 5.1; 0.05x on Opus 5.5 and Sonnet 5.5)

Storage cost
Not stated
How to enable

Add cache_control at the top level of the request (automatic) or on individual content blocks; add "ttl": "1h" for the 1-hour cache.

Request fragment
"cache_control": {"type": "ephemeral", "ttl": "1h"}
From the docs
“Automatic caching: Add a single `cache_control` field at the top level of your request. The system automatically applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow.”
“Explicit cache breakpoints: Place `cache_control` directly on individual content blocks for fine-grained control over exactly what gets cached.”
“By default, the cache has a 5-minute lifetime. The cache is refreshed for no additional cost each time the cached content is used.”
“By default, automatic caching uses a 5-minute TTL. You can specify a 1-hour TTL at 2x the base input token price:”
“512 tokens for Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5.5, Claude Fable 5, Claude Mythos 5, and Claude Haiku 5.5”
“Shorter prompts cannot be cached, even if marked with `cache_control`. Any requests to cache fewer than this number of tokens will be processed without caching, and no error is returned.”
“5-minute cache write tokens are 1.25 times the base input tokens price”
“1-hour cache write tokens are 2 times the base input tokens price”
“Cache read (hit) | 0.1x base input price (0.025x on Claude Fable 5.1 and Claude Mythos 5.1; 0.05x on Claude Opus 5.5 and Claude Sonnet 5.5)”
“caching pays off after one cache read for the 5-minute duration (1.25x write), or after two cache reads for the 1-hour duration (2x write).”
“You can define up to 4 cache breakpoints”
“These multipliers stack with other pricing modifiers, including the Batch API discount and data residency.”

Sources: platform.claude.com/docs/en/build-with-claude/prompt-caching, platform.claude.com/docs/en/about-claude/pricing

Batch API

Discount

50% off

50% off (all usage at 50% of standard prices)

Window

Most batches finish in under 1 hour; batches expire if not complete within 24 hours; results available 29 days

Limits

100,000 Message requests or 256 MB per batch, whichever is reached first

Endpoint

POST /v1/messages/batches (Message Batches API)

Stacks with caching

Yes: the batch discount and the cache discount combine

“Batch API and prompt caching discounts can be combined.”
platform.claude.com
From the docs
“The Batches API offers significant cost savings. All usage is charged at 50% of the standard API prices.”
“The system processes each batch as fast as possible, with most batches completing within 1 hour. You can access batch results when all messages have completed or after 24 hours, whichever comes first. Batches expire if processing does not complete within 24 hours.”
“A Message Batch is limited to either 100,000 Message requests or 256 MB in size, whichever is reached first.”
“Batch results are available for 29 days after creation.”
“Batch reached its 24-hour expiration before this request could be sent to the model. You will not be billed for these requests.”

Sources: platform.claude.com/docs/en/build-with-claude/batch-processing, platform.claude.com/docs/en/about-claude/pricing

Service tiers

Flex trades speed or availability for a lower price; Priority and Fast cost more for lower latency or reserved capacity.

Fast

Fast mode

Set with speed: "fast"

Price effect

Premium pricing: 2x standard on supported models (Opus 5.5 $8/$40 vs $4/$20; Opus 5 and Opus 4.8 $10/$50 vs $5/$25)

Derived from the pricing table. The pages say only "priced at a multiplier on standard rates". The 2x is computed from the fast mode pricing table against the model pricing table: Opus 5.5 $8/$40 vs $4/$20; Opus 5 / Opus 4.8 $10/$50 vs $5/$25.

Trade-off

Up to 2.5x higher output tokens per second (OTPS, not TTFT); dedicated rate limit; not available with Batch API

From the docs
“Fast mode delivers up to 2.5x higher output tokens per second from Claude Opus 5.5, Claude Opus 5, and Claude Opus 4.8 at premium pricing. Set `speed: "fast"` on your request to opt in.”
“Speed benefits are focused on output tokens per second (OTPS), not time to first token (TTFT)”
“Fast mode is priced at a multiplier on standard rates across the full context window, including requests over 200k input tokens.”
“Fast mode is not available with the Batch API.”

Sources: platform.claude.com/docs/en/build-with-claude/fast-mode, platform.claude.com/docs/en/about-claude/pricing

Priority

Priority Tier

Set with service_tier: "auto"

Price effect

Capacity commitment, no list price; new commitments are no longer sold, so only organizations with an existing commitment can use it (through their contract end date)

Trade-off

Prioritized over all other requests to minimize overloaded errors

From the docs
“Priority Tier capacity commitments are no longer available for purchase.”
platform.claude.com
“Available only to organizations with an existing capacity commitment”
platform.claude.com

Other discounts and surcharges

  • Volume discounts (negotiated)

    “Volume discounts may be available for high-volume users. These are negotiated on a case-by-case basis.”
    platform.claude.com

  • Batch + caching stack

    “Batch API and prompt caching discounts can be combined.”
    platform.claude.com

  • Surcharge to avoid: US-only inference (inference_geo: "us") 1.1x

    “For Claude 4.6 and later models, specifying US-only inference through the `inference_geo` parameter incurs a 1.1x multiplier on all token pricing categories, including input tokens, output tokens, cache writes, and cache reads.”
    platform.claude.com

Anthropic per-model prices

USD per 1M tokens from the provider's own pricing page, newest release first. A dash means the model has no such rate. Muted rows are scheduled for shutdown or deprecated.

Anthropic models with input, cache, batch, output and fast mode prices per 1M tokens
ModelInputCached inputCache write1-hour writeBatch inputOutputBatch outputFast modeinput / output
Claude Haiku 5.5$0.10$0.0190% off$0.125$0.20$0.05$0.50$0.25-
Claude Sonnet 5.5$2$0.1095% off$2.50$4$1$10$5-
Claude Opus 5.5$4$0.2095% off$5$8$2$20$10$8 / $40
Claude Fable 5.1$10$0.2597.5% off$12.50$20$5$50$25-
Claude Mythos 5.1$10$0.2597.5% off$12.50$20$5$50$25-
Claude Opus 5$5$0.5090% off$6.25$10$2.50$25$12.50$10 / $50
Claude Sonnet 5$2$0.2090% off$2.50$4$1$10$5-
Claude Fable 5$10$190% off$12.50$20$5$50$25-
Claude Mythos 5$10$190% off$12.50$20$5$50$25-
Claude Opus 4.8$5$0.5090% off$6.25$10$2.50$25$12.50$10 / $50
Claude Opus 4.7$5$0.5090% off$6.25$10$2.50$25$12.50-
Claude Sonnet 4.6$3$0.3090% off$3.75$6$1.50$15$7.50-
Claude Opus 4.6$5$0.5090% off$6.25$10$2.50$25$12.50-
Claude Opus 4.5$5$0.5090% off$6.25$10$2.50$25$12.50-
Claude Haiku 4.5$1$0.1090% off$1.25$2$0.50$5$2.50-
Claude Sonnet 4.5Scheduled for shutdown or deprecated$3$0.3090% off$3.75$6$1.50$15$7.50-

Research notes

docs.claude.com redirects to platform.claude.com. Fast mode 2x is derived from the pricing tables.

FAQ

Does Anthropic cache prompts automatically?

Explicit: nothing is cached until you mark it. Opt-in via cache_control: either one top-level cache_control (automatic breakpoint placement) or explicit block-level breakpoints (up to 4). No caching without cache_control. How to enable: Add cache_control at the top level of the request (automatic) or on individual content blocks; add "ttl": "1h" for the 1-hour cache.

How much does a cache hit cost on Anthropic?

Across 15 current models with a cache read price, a hit saves 90% (90% to 97.5%) off the input price. The docs say: 0.1x base input (0.025x on Fable 5.1 and Mythos 5.1; 0.05x on Opus 5.5 and Sonnet 5.5).

Does Anthropic have a batch API discount?

50% off (all usage at 50% of standard prices). Window: Most batches finish in under 1 hour; batches expire if not complete within 24 hours; results available 29 days. Stacks with caching: yes: the batch discount and the cache discount combine.

What service tiers does Anthropic offer?

Fast mode: Premium pricing: 2x standard on supported models (Opus 5.5 $8/$40 vs $4/$20; Opus 5 and Opus 4.8 $10/$50 vs $5/$25) (derived from the pricing table). Priority Tier: Capacity commitment, no list price; new commitments are no longer sold, so only organizations with an existing commitment can use it (through their contract end date).

Sources

Last verified from Anthropic (Claude API)'s own pages below. Quotes are verbatim; each was matched against the live page.

Newsletter

Know when a provider changes its prices

One email per week. New model releases, pricing changes and what they mean for what you build.