Prompt Cache ROI · for API builders

Does prompt caching actually save you money?

Cache economics vary sharply by provider. Anthropic charges a 25% write premium - so caching only saves if your hit rate clears break-even. OpenAI and Google charge no write premium - caching is always free upside.

Pricing verified: 2026-07-28 94 cache-capable models Break-even analysis

What this calculator does

Does prompt caching actually save you money on your workload — or does it cost you more?

Why use it

Anthropic charges a 25% write premium — caching can LOSE money below a 22% hit rate
OpenAI and Google have no write premium — caching is pure upside from hit rate 0.01% upward
See exact break-even hit rate for your chosen model before enabling cache in production
Compare cache ROI across all 8 cache-capable models at one click

New to this calculator? Start with the ⚡ Playground — a few sliders, instant ballpark. Then switch to the 🧮 Calculator for your exact number.

Two ways to use this: visualize in the Playground, then get your number in the Calculator.

▶Playground A quick, visual way to see which factors move your savings the most. Open the playground → ▤Calculator Enter your real workload for a precise savings you can apply to your own usage. Go to the calculator →

Prompt caching — is it worth turning on?

Caching reused prompt prefixes (system prompts, long context) earns a steep discount on the cached tokens. Worth it when reuse is high.

Cheaper option

—

Change the input sliders below to compare.

Input spend / mo 2,400

Cache hit rate 0.5

Cached discount 90

No cache

—

/ month

With cache

—

/ month

💡Cheaper is pure cost. Caching wins whenever enough of your input repeats — there’s rarely a downside beyond a little engineering.

Prompt caching only pays off above a certain reuse rate. Enter your usage below to see whether it actually saves you money.

📊 Not sure of a value? Fields with a ▾ Typical pill offer broad industry ballparks (sourced typical ranges) so you can move forward now — your result gets more accurate as you replace them with your own measured numbers. Values marked * are rough estimates.

🎮 Interactive Guide & Calculator Playground →

📥 Inputs you provide

ModelPick a cache-capable model
Requests per monthTotal monthly call volume
Input tokens / requestAverage input size
Output tokens / requestAverage output size
Cacheable portion of inputStatic-prefix share of input
Cache hit rateHow often the cache is warm

📤 Outputs you get

No-cache costMonthly cost with caching off
With-cache costMonthly cost with caching on
Monthly savingsDollars saved (or lost) per month
Break-even hit rateMinimum hit rate to save

🎯 Use your results to

🎯

Find your break-even

See the exact hit rate caching must clear before you flip it on in production

💾

Quantify the savings

Real monthly and annual dollars at your hit rate — not a vendor headline number

🆚

Compare providers

Anthropic vs OpenAI vs Google cache economics side-by-side on the same workload

🔌

Integrate with your agents

MCP available so agentic workflows can pull cache ROI programmatically

Enter key values

Pick your model, set requests/month, and move the two sliders: cacheable % (how much of your prompt is reusable) and hit rate % (how often the cache is still warm).

Pick a workload preset

RAG, long-prompt, doc Q&A, coding, agent loop, or one-off chat. Each preset sets both sliders to real-world averages.

Read the break-even bar

If your YOU marker is RIGHT of the threshold = you save money. LEFT of it = you pay more with cache on.

Act on it

If you're below break-even on Anthropic, either raise hit rate (longer TTL, consolidate calls) or disable cache. If you're above, no changes needed.

Enter every field

Model determines the cache economics — Anthropic's read discount is 90% but comes with a 25% write premium. OpenAI's 50% read discount has no write premium. Requests/month scales total impact. Input tokens/request matters because cache savings only apply to input; heavy-output workloads see smaller % savings. Output tokens/request scales the uncachable portion of cost.

Tune with cacheable/hit sliders

Cacheable % = the static prefix fraction of your prompt (system rules, few-shot examples, retrieved docs). If your system prompt is 800 tokens and query is 200, cacheable = 80%. Hit rate % = how often the cache is still warm when a new request arrives. Depends on TTL (5 min Anthropic, 7 min OpenAI, 1 hour Gemini) vs time-between-calls. High cacheable + low hit rate = cache is expiring before reuse — bump request cadence or use extended TTL.

Interpret the 3 cards + break-even bar

Left card = cost without cache. Middle card = cost with cache (based on your sliders). Right card = monthly savings (flips red if cache is costing you money). The break-even bar shows the minimum hit rate needed to save; YOU marker shows where you actually are. Model comparison table at bottom stacks all 8 cache-capable models side-by-side — the winner column often surprises.

Next actions

Stack cache savings with routing and batch API for compound savings (often 80%+ total). If your use case has long-lived context (agent loops, stable RAG docs), consider Anthropic's 1-hour extended cache (2× write cost but dramatically more hits). Always verify savings hold at production scale — first-request-after-deploy always misses.

👇 Now try the calculator below with your own AI workloads

🎛 CALCULATOR

⚙️ Your workload

Start with a preset, then adjust.

Model Hint: Which model you run. Caching savings depend on its rates.

Requests per month Hint: Requests per month.

Input tokens / request Hint: Words sent in per request.

Output tokens / request Hint: Words sent back per request.

Cacheable portion of input

60%

Hint: Share of your input that repeats every time (system prompt, docs, examples).

Cache hit rate

70%

Hint: Share of requests arriving while the cache is still warm. Higher = more savings.

📈 RESULTS

No cache

With cache

Monthly savings

Break-even hit rate - your hit rate must exceed this to save

22%

0% (caching always loses) 100% (every request hits)

💡 Recommendations

📋 Cache ROI across all 94 cache-capable models

Same workload, different models. Green row = biggest savings, gold = your current model.

Model	Base / cached input	No cache	With cache	Savings

RAG Pipeline Cost → Generic LLM cost calculator → Agent loop cost → Get an AI cost audit →

📋 What now?

If you're above break-even — ship caching on the static prefix first (system prompt + tool definitions + few-shot examples). That's ~80% of the savings for ~20% of the effort.
If you're below break-even on Anthropic — raise the hit rate (consolidate calls, use the 1-hour extended cache) or move that workload to a no-write-premium provider where any hit rate saves.
Verify before you bank it — measure the real hit rate in production for 1-2 weeks; the first request after each TTL expiry always pays full price, so steady-state numbers differ from the estimate.

📅 Book a working session to apply this to your workload →

Need help using this calculator for your workloads?

AICost.ai has 50+ calculators and playbooks. Schedule an AvatarVA meeting and we'll work through your real cost scenarios across AI & Cloud: visibility, cost reduction, optimization, forecasting and capacity planning, without sacrificing accuracy or performance.

📅 Schedule an AvatarVA meeting →

Vendor / Model	Field	Why it’s inferred
Anthropic — Claude Sonnet 4.6	`cachedInput`	Derived at 10% of input rate — Anthropic publishes 90% cache-hit discount on this tier.
Anthropic — Claude Sonnet 4.5	`cachedInput`	Derived at 10% of input rate; same 90% cache-hit convention as Sonnet 4.6.
Anthropic — Claude Sonnet 4.5	`batchInput`	Derived at 50% of standard input — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Sonnet 4.5	`batchOutput`	Derived at 50% of standard output — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Haiku 4.5	`cachedInput`	Derived at 10% of input rate — Anthropic 90% cache-hit discount convention.
OpenAI — GPT-5.4 Mini	`cachedInput`	Derived at 10% of input — OpenAI documents automatic 90% discount on cache hits across GPT-5.x tier.
OpenAI — GPT-5.4 Nano	`cachedInput`	Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Nano	`batchInput`	Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Nano	`batchOutput`	Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro	`cachedInput`	Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Pro	`batchInput`	Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro	`batchOutput`	Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.2	`cachedInput`	Derived at 10% of input; no residency uplift.
OpenAI — GPT-5.2	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5.2	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5	`cachedInput`	Derived at 10% of input.
OpenAI — GPT-5	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5.5 Pro	`cachedInput`	Derived at 10% of input — OpenAI does not publish a cached rate for *-pro models; using the family convention.
OpenAI — GPT-5.5 Pro	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5.5 Pro	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5.2 Pro	`cachedInput`	Derived at 10% of input — pro-tier convention.
OpenAI — GPT-5.2 Pro	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5.2 Pro	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5.1	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5.1	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5 Pro	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5 Pro	`batchOutput`	Derived at 50% of output.
OpenAI — GPT-5 Nano	`cachedInput`	Derived at 10% of input.
OpenAI — GPT-5 Nano	`batchInput`	Derived at 50% of input.
OpenAI — GPT-5 Nano	`batchOutput`	Derived at 50% of output.
Google — Gemini 3 Flash	`cachedInput`	Derived at 10% of input — Google caching discount convention ~90%.
Google — Gemini 3.1 Flash-Lite	`cachedInput`	Derived at 10% of input — Google caching convention.
Google — Gemini 3.1 Flash-Lite	`batchInput`	Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 3.1 Flash-Lite	`batchOutput`	Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Pro	`cachedInput`	Derived at 10% of input.
Google — Gemini 2.5 Flash	`cachedInput`	Derived at 10% of input.
Google — Gemini 2.5 Flash-Lite	`cachedInput`	Derived at 10% of input — Google caching convention.
Google — Gemini 2.5 Flash-Lite	`batchInput`	Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Flash-Lite	`batchOutput`	Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash	`cachedInput`	Derived at 25% of input per Google 2.0 family caching rates.
Google — Gemini 2.0 Flash	`batchInput`	Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash	`batchOutput`	Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash-Lite	`cachedInput`	Derived at 10% of input — Google caching convention.
Google — Gemini 2.0 Flash-Lite	`batchInput`	Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash-Lite	`batchOutput`	Derived at 50% of output — Google Batch API uniform 50% discount.
xAI — Grok 4 (legacy)	`cachedInput`	Extrapolated at 25% of base.

Vendor / Model

Field

Why it’s inferred

Anthropic — Claude Sonnet 4.6

cachedInput

Derived at 10% of input rate — Anthropic publishes 90% cache-hit discount on this tier.

Anthropic — Claude Sonnet 4.5

cachedInput

Derived at 10% of input rate; same 90% cache-hit convention as Sonnet 4.6.

Anthropic — Claude Sonnet 4.5

batchInput

Derived at 50% of standard input — Anthropic documents uniform 50% Batch discount.

Anthropic — Claude Sonnet 4.5

batchOutput

Derived at 50% of standard output — Anthropic documents uniform 50% Batch discount.

Anthropic — Claude Haiku 4.5

cachedInput

Derived at 10% of input rate — Anthropic 90% cache-hit discount convention.

OpenAI — GPT-5.4 Mini

cachedInput

Derived at 10% of input — OpenAI documents automatic 90% discount on cache hits across GPT-5.x tier.

OpenAI — GPT-5.4 Nano

cachedInput

Derived at 10% of input — OpenAI 90% cache-hit convention.

OpenAI — GPT-5.4 Nano

batchInput

Derived at 50% of input — OpenAI Batch API uniform 50% discount.

OpenAI — GPT-5.4 Nano

batchOutput

Derived at 50% of output — OpenAI Batch API uniform 50% discount.

OpenAI — GPT-5.4 Pro

cachedInput

Derived at 10% of input — OpenAI 90% cache-hit convention.

OpenAI — GPT-5.4 Pro

batchInput

Derived at 50% of input — OpenAI Batch API uniform 50% discount.

OpenAI — GPT-5.4 Pro

batchOutput

Derived at 50% of output — OpenAI Batch API uniform 50% discount.

OpenAI — GPT-5.2

cachedInput

Derived at 10% of input; no residency uplift.

OpenAI — GPT-5.2

batchInput

Derived at 50% of input.

OpenAI — GPT-5.2

batchOutput

Derived at 50% of output.

OpenAI — GPT-5

cachedInput

Derived at 10% of input.

OpenAI — GPT-5

batchInput

Derived at 50% of input.

OpenAI — GPT-5

batchOutput

Derived at 50% of output.

OpenAI — GPT-5.5 Pro

cachedInput

Derived at 10% of input — OpenAI does not publish a cached rate for *-pro models; using the family convention.

OpenAI — GPT-5.5 Pro

batchInput

Derived at 50% of input.

OpenAI — GPT-5.5 Pro

batchOutput

Derived at 50% of output.

OpenAI — GPT-5.2 Pro

cachedInput

Derived at 10% of input — pro-tier convention.

OpenAI — GPT-5.2 Pro

batchInput

Derived at 50% of input.

OpenAI — GPT-5.2 Pro

batchOutput

Derived at 50% of output.

OpenAI — GPT-5.1

batchInput

Derived at 50% of input.

OpenAI — GPT-5.1

batchOutput

Derived at 50% of output.

OpenAI — GPT-5 Pro

batchInput

Derived at 50% of input.

OpenAI — GPT-5 Pro

batchOutput

Derived at 50% of output.

OpenAI — GPT-5 Nano

cachedInput

Derived at 10% of input.

OpenAI — GPT-5 Nano

batchInput

Derived at 50% of input.

OpenAI — GPT-5 Nano

batchOutput

Derived at 50% of output.

Google — Gemini 3 Flash

cachedInput

Derived at 10% of input — Google caching discount convention ~90%.

Google — Gemini 3.1 Flash-Lite

cachedInput

Derived at 10% of input — Google caching convention.

Google — Gemini 3.1 Flash-Lite

batchInput

Derived at 50% of input — Google Batch API uniform 50% discount.

Google — Gemini 3.1 Flash-Lite

batchOutput

Derived at 50% of output — Google Batch API uniform 50% discount.

Google — Gemini 2.5 Pro

cachedInput

Derived at 10% of input.

Google — Gemini 2.5 Flash

cachedInput

Derived at 10% of input.

Google — Gemini 2.5 Flash-Lite

cachedInput

Derived at 10% of input — Google caching convention.

Google — Gemini 2.5 Flash-Lite

batchInput

Derived at 50% of input — Google Batch API uniform 50% discount.

Google — Gemini 2.5 Flash-Lite

batchOutput

Derived at 50% of output — Google Batch API uniform 50% discount.

Google — Gemini 2.0 Flash

cachedInput

Derived at 25% of input per Google 2.0 family caching rates.

Google — Gemini 2.0 Flash

batchInput

Derived at 50% of input — Google Batch API uniform 50% discount.

Google — Gemini 2.0 Flash

batchOutput

Derived at 50% of output — Google Batch API uniform 50% discount.

Google — Gemini 2.0 Flash-Lite

cachedInput

Derived at 10% of input — Google caching convention.

Google — Gemini 2.0 Flash-Lite

batchInput

Derived at 50% of input — Google Batch API uniform 50% discount.

Google — Gemini 2.0 Flash-Lite

batchOutput

Derived at 50% of output — Google Batch API uniform 50% discount.

xAI — Grok 4 (legacy)

cachedInput

Extrapolated at 25% of base.

Does prompt caching actually save you money?

Prompt Cache ROI Calculator

Results

Go deeper

Need help using this calculator for your workloads?

Methodology

Primary sources

Inferred values (marked with * in calculator tables)

Immediate steps you can take