API pricing

Prompt caching and batch API prices

Reading a prompt from the cache costs 2 % to 25 % of the normal input price for the 18 models whose provider lists a cache price; the OpenRouter list also carries cache-read prices for 209 models, shown on their model cards and used by the model picker; on OpenRouter, 61 of 79 batch listings cost exactly half the standard price, and 7 cost more on at least one side (prices as listed in the United States on ).

Cache prices on provider price lists

Task receipt: Direct API / OpenRouter: $0.348; Warm cache: $0.082; Published batch: $0.174 Same price on both channels Calculated · Document analysis Assumption: 3 steps; initial input 50,500 tokens; growth 2,500 tokens/step; visible output 1,000 tokens/step. Month: 14,600 tasks. No hidden reasoning or retries.
Task receipt.Same price on both channels Calculated · Document analysis Assumption: 3 steps; initial input 50,500 tokens; growth 2,500 tokens/step; visible output 1,000 tokens/step. Month: 14,600 tasks. No hidden reasoning or retries. Open image · Explore the chart · Measured Lab · CC BY 4.0
ModelInput, per 1MCache read, per 1MCache write, per 1MCache storage, per 1M
GPT-6 AstraOpenAI$10.00$1.0010 % of input$12.50×1.25 inputUnknownNot listed
GPT-6 SolOpenAI$2.00$0.2010 % of input$2.50×1.25 inputUnknownNot listed
GPT-6.1 SolOpenAI$2.00$0.105 % of input$2.50×1.25 inputUnknownNot listed
GPT-6 LunaOpenAI$0.10$0.0110 % of input$0.125×1.25 inputUnknownNot listed
Claude Fable 5.1Anthropic$10.00$0.252.5 % of input$12.50 5 min$20.00 1 hourUnknownNot listed
Claude Opus 5.5Anthropic$4.00$0.205 % of input$5.00 5 min$8.00 1 hourUnknownNot listed
Claude Sonnet 5.5Anthropic$2.00$0.105 % of input$2.50 5 min$4.00 1 hourUnknownNot listed
Claude Haiku 5.5Anthropic$0.10$0.0110 % of input$0.125 5 min$0.20 1 hourUnknownNot listed
Gemini 3.1 Pro PreviewGoogle$2.00$0.2010 % of inputUnknownNot listed$4.50 per hour
Gemini 3.8 FlashGoogle$0.75$0.07510 % of inputUnknownNot listed$0.50 per hour
DeepSeek V4 Pro 0813DeepSeek$1.32$0.0443.3 % of inputUnknownNot listedUnknownNot listed
DeepSeek V4.1 Flash 1017DeepSeek$0.30$0.0062 % of inputUnknownNot listedUnknownNot listed
Grok 4.7xAI$2.00$0.5025 % of inputUnknownNot listedUnknownNot listed
Mistral Medium 3.5Mistral$1.50$0.1510 % of inputUnknownNot listedUnknownNot listed
Mistral Large 4Mistral$0.70$0.0710 % of inputUnknownNot listedUnknownNot listed
Mistral Large 3 2512Mistral$0.50$0.0510 % of inputUnknownNot listedUnknownNot listed
Kimi K3MoonshotAI$3.00$0.3010 % of inputUnknownNot listedUnknownNot listed
GLM 5.3Z.ai$1.40$0.2618.6 % of inputUnknownNot listedUnknownNot listed

Published Provider price lists, prices as listed in the United States on . “Not listed”: the price list does not publish that price. Anthropic lists one write price for a 5-minute cache and one for a 1-hour cache.

What the cache does to a task

ModelL2, no cacheL2, 75 % cachedChangeL3, no cacheL3, 85 % cachedChange
GPT-6 AstraOpenAI$1.74$0.480−72.4 %$13.61$2.87−78.9 %
GPT-6 SolOpenAI$0.348$0.096−72.4 %$2.72$0.574−78.9 %
GPT-6.1 SolOpenAI$0.348$0.082−76.5 %$2.72$0.452−83.4 %
GPT-6 LunaOpenAI$0.017$0.0048−72.4 %$0.136$0.029−78.9 %
Claude Fable 5.1Anthropic$1.74$0.374−78.5 %$13.61$1.96−85.6 %
Claude Opus 5.5Anthropic$0.696$0.164−76.5 %$5.44$0.904−83.4 %
Claude Sonnet 5.5Anthropic$0.348$0.082−76.5 %$2.72$0.452−83.4 %
Claude Haiku 5.5Anthropic$0.017$0.0048−72.4 %$0.424$0.082−80.7 %
Gemini 3.1 Pro PreviewGoogle$0.354$0.100−71.9 %$2.74$0.546−80.1 %
Gemini 3.8 FlashGoogle$0.131$0.035−73.1 %$1.02$0.198−80.6 %
DeepSeek V4 Pro 0813DeepSeek$0.222$0.041−81.3 %$1.77$0.218−87.7 %
DeepSeek V4.1 Flash 1017DeepSeek$0.051$0.0097−81.0 %$0.405$0.047−88.3 %
Grok 4.7xAI$0.336$0.124−63.1 %$2.69$0.858−68.1 %
Mistral Medium 3.5Mistral$0.261$0.070−73.1 %$2.04$0.396−80.6 %
Mistral Large 4Mistral$0.118$0.029−75.7 %$0.940$0.172−81.7 %
Mistral Large 3 2512Mistral$0.084$0.020−75.7 %$0.671$0.123−81.7 %
Kimi K3MoonshotAI$0.522$0.140−73.1 %$4.08$0.792−80.6 %
GLM 5.3Z.ai$0.236$0.075−68.3 %$1.88$0.493−73.8 %

Calculated The same tasks as in the task cost calculator, 20 tasks an hour, so a 5-minute cache stays warm between tasks. With cache, the cached share of the step 1 input is read at the cache-read price; from step 2 the whole earlier context is read from cache and only the new tokens of the step are written at the cache-write price (at the input price where no write price is listed), as providers bill a loop that resends its history. Storage fees are not included.

Batch prices on OpenRouter 79 models

ModelStandard, in / outBatch, in / outBatch input vs standardBatch output vs standard
Claude Fable 5Anthropic$10.00 / $50.00$5.00 / $25.00−50.0 %−50.0 %
Claude Fable 5.1Anthropic$10.00 / $50.00$5.00 / $25.00−50.0 %−50.0 %
Claude Haiku 4.5Anthropic$1.00 / $5.00$0.50 / $2.50−50.0 %−50.0 %
Claude Haiku 5.5Anthropic$0.10 / $0.50$0.05 / $0.25−50.0 %−50.0 %
Claude Opus 4.1Anthropic$15.00 / $75.00$7.50 / $37.50−50.0 %−50.0 %
Claude Opus 4.5Anthropic$5.00 / $25.00$2.50 / $12.50−50.0 %−50.0 %
Claude Opus 4.6Anthropic$5.00 / $25.00$2.50 / $12.50−50.0 %−50.0 %
Claude Opus 4.7Anthropic$5.00 / $25.00$2.50 / $12.50−50.0 %−50.0 %
Claude Opus 4.8Anthropic$5.00 / $25.00$2.50 / $12.50−50.0 %−50.0 %
Claude Opus 5Anthropic$5.00 / $25.00$2.50 / $12.50−50.0 %−50.0 %
Claude Opus 5.5Anthropic$4.00 / $20.00$2.00 / $10.00−50.0 %−50.0 %
Claude Sonnet 4.5Anthropic$3.00 / $15.00$1.50 / $7.50−50.0 %−50.0 %
Claude Sonnet 4.6Anthropic$3.00 / $15.00$1.50 / $7.50−50.0 %−50.0 %
Claude Sonnet 5Anthropic$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
Claude Sonnet 5.5Anthropic$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
DeepSeek V4 Flash 0731DeepSeek$0.0434 / $0.30$0.048 / $0.144+10.6 %−52.0 %
DeepSeek V4.1 Flash 0910DeepSeek$0.051 / $0.567$0.011 / $0.20−78.4 %−64.7 %
DeepSeek V4.1 Flash 1017DeepSeek$0.006 / $0.50$0.011 / $0.20+83.3 %−60.0 %
Gemini 3 Flash PreviewGoogle$0.50 / $3.00$0.25 / $1.50−50.0 %−50.0 %
Gemini 3.1 Flash LiteGoogle$0.25 / $1.50$0.125 / $0.75−50.0 %−50.0 %
Gemini 3.1 Pro PreviewGoogle$2.00 / $12.00$1.00 / $6.00−50.0 %−50.0 %
Gemini 3.5 FlashGoogle$1.50 / $9.00$0.75 / $4.50−50.0 %−50.0 %
Gemini 3.5 Flash LiteGoogle$0.30 / $2.50$0.15 / $1.25−50.0 %−50.0 %
Gemini 3.6 FlashGoogle$0.75 / $3.75$0.375 / $1.875−50.0 %−50.0 %
Gemini 3.7 FlashGoogle$0.75 / $3.75$0.375 / $1.875−50.0 %−50.0 %
Gemini 3.8 FlashGoogle$0.825 / $4.125$0.375 / $1.875−54.5 %−54.5 %
Gemini 4 ArgonGoogle$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
Muse Glimmer 30BMeta$0.30 / $1.20$0.24 / $0.96−20.0 %−20.0 %
MiniMax M3MiniMax$0.30 / $1.20$0.224 / $0.88−25.3 %−26.7 %
Mistral Medium 3.5Mistral$1.50 / $7.50$0.75 / $3.75−50.0 %−50.0 %
Kimi K3MoonshotAI$0.192 / $16.00$0.917 / $6.00+377.6 %−62.5 %
Morph V3 FastMorph$0.80 / $1.20$0.40 / $0.60−50.0 %−50.0 %
Morph V3 LargeMorph$0.90 / $1.90$0.45 / $0.95−50.0 %−50.0 %
GPT-4.1OpenAI$2.00 / $8.00$1.00 / $4.00−50.0 %−50.0 %
GPT-4.1 MiniOpenAI$0.40 / $1.60$0.20 / $0.80−50.0 %−50.0 %
GPT-4.1 NanoOpenAI$0.10 / $0.40$0.05 / $0.20−50.0 %−50.0 %
GPT-4oOpenAI$2.50 / $10.00$1.25 / $5.00−50.0 %−50.0 %
GPT-4o-miniOpenAI$0.15 / $0.60$0.075 / $0.30−50.0 %−50.0 %
GPT-5OpenAI$1.25 / $10.00$0.625 / $5.00−50.0 %−50.0 %
GPT-5 MiniOpenAI$0.25 / $2.00$0.125 / $1.00−50.0 %−50.0 %
GPT-5 NanoOpenAI$0.05 / $0.40$0.025 / $0.20−50.0 %−50.0 %
GPT-5 ProOpenAI$15.00 / $120.00$7.50 / $60.00−50.0 %−50.0 %
GPT-5.1OpenAI$1.25 / $10.00$0.625 / $5.00−50.0 %−50.0 %
GPT-5.2OpenAI$1.75 / $14.00$0.875 / $7.00−50.0 %−50.0 %
GPT-5.2 ProOpenAI$21.00 / $168.00$10.50 / $84.00−50.0 %−50.0 %
GPT-5.4OpenAI$2.50 / $15.00$1.25 / $7.50−50.0 %−50.0 %
GPT-5.4 MiniOpenAI$0.75 / $4.50$0.375 / $2.25−50.0 %−50.0 %
GPT-5.4 NanoOpenAI$0.20 / $1.25$0.10 / $0.625−50.0 %−50.0 %
GPT-5.4 ProOpenAI$30.00 / $180.00$15.00 / $90.00−50.0 %−50.0 %
GPT-5.5OpenAI$5.00 / $30.00$2.50 / $15.00−50.0 %−50.0 %
GPT-5.5 ProOpenAI$30.00 / $180.00$15.00 / $90.00−50.0 %−50.0 %
GPT-5.6 LunaOpenAI$0.20 / $1.20$0.10 / $0.60−50.0 %−50.0 %
GPT-5.6 Luna ProOpenAI$0.20 / $1.20$0.10 / $0.60−50.0 %−50.0 %
GPT-5.6 SolOpenAI$4.00 / $20.00$2.00 / $10.00−50.0 %−50.0 %
GPT-5.6 Sol ProOpenAI$4.00 / $20.00$2.00 / $10.00−50.0 %−50.0 %
GPT-5.6 TerraOpenAI$2.00 / $12.00$1.00 / $6.00−50.0 %−50.0 %
GPT-5.6 Terra ProOpenAI$2.00 / $12.00$1.00 / $6.00−50.0 %−50.0 %
GPT-6 AstraOpenAI$10.00 / $50.00$5.00 / $25.00−50.0 %−50.0 %
GPT-6 Astra ProOpenAI$10.00 / $50.00$5.00 / $25.00−50.0 %−50.0 %
GPT-6 LunaOpenAI$0.10 / $0.50$0.05 / $0.25−50.0 %−50.0 %
GPT-6 Luna ProOpenAI$0.10 / $0.50$0.05 / $0.25−50.0 %−50.0 %
GPT-6 SolOpenAI$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
GPT-6 Sol ProOpenAI$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
GPT-6.1 SolOpenAI$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
GPT-6.1 Sol ProOpenAI$2.00 / $10.00$1.00 / $5.00−50.0 %−50.0 %
GPT-6.1 TerraOpenAI$1.00 / $5.00$0.50 / $2.50−50.0 %−50.0 %
GPT-6.1 Terra ProOpenAI$1.00 / $5.00$0.50 / $2.50−50.0 %−50.0 %
gpt-oss-120bOpenAI$0.15 / $0.60$0.0296 / $0.136−80.3 %−77.3 %
gpt-oss-20bOpenAI$0.03 / $0.13$0.024 / $0.112−20.0 %−13.8 %
o3OpenAI$2.00 / $8.00$1.00 / $4.00−50.0 %−50.0 %
Qwen3.7 PlusQwen$0.32 / $1.28$0.23 / $0.92−28.1 %−28.1 %
Qwen3.8 27BQwen$0.15 / $1.875$0.16 / $2.00+6.7 %+6.7 %
Inkling SmallThinking Machines$0.45 / $1.20$0.36 / $0.96−20.0 %−20.0 %
Grok 4.3xAI$1.25 / $2.50$1.00 / $2.00−20.0 %−20.0 %
Grok 4.6xAI$2.00 / $6.00$1.60 / $4.80−20.0 %−20.0 %
Grok 4.7xAI$2.00 / $6.00$1.60 / $4.80−20.0 %−20.0 %
GLM 5.2Z.ai$0.006 / $6.80$0.623 / $1.271+10283.3 %−81.3 %
GLM 5.3Z.ai$0.0699 / $1.40$0.45 / $2.00+543.8 %+42.9 %
GLM 5.3 FlashZ.ai$0.0335 / $0.083$0.05 / $0.18+49.3 %+116.9 %

Published OpenRouter models list, prices as listed in the United States on · USD per 1M tokens. Standard: the model’s own listing; batch: its batch listing. A batch listing can be served by a different provider than the cheapest standard one.

All API prices · Cheapest per task · Long-context cliff · Rate limits by tier · What breaks when you switch · Rate limit planner · Task cost calculator · All models on OpenRouter · How we check prices

Explore the numbers

Claude Sonnet 5.5

Same price on both channels

Direct API / OpenRouter$0.348Warm cache$0.082Published batch$0.174$0.00$0.10$0.20$0.30$0.40Direct API / OpenRouter$0.348Warm cache$0.082Published batch$0.174$0.00$0.10$0.20$0.30$0.40

Calculated · Document analysis

Assumption: 3 steps; initial input 50,500 tokens; growth 2,500 tokens/step; visible output 1,000 tokens/step. Month: 14,600 tasks. No hidden reasoning or retries.

Warm-cache saving: $0.266 per task. Dashed outline: the no-cache bill.

Assumption: first-step cache fraction 75%; earlier context is read again and new context is written. All calls stay within the cache lifetime.

Formula: sum of input, output, cache reads and cache writes × their published rates / 1M. Each step uses its own context tier. Missing cache or batch prices are Unknown.

Fresh input · Visible output · Cache read · Cache write

Claude Sonnet 5.5

5-minute cache: pays back from 2 uses.

1-hour cache: pays back from 3 uses.

USD · Calculated$0.00$5.00$10.00$15.00$20.00123510Uses of the same prefixUSD · Calculated$0.00$5.00$10.00$15.00$20.00123510Uses of the same prefix

Fresh input · 5-minute cache · 1-hour cache

UsesFresh input5-minute cache1-hour cache
1$2.00$2.50$4.00
2$4.00$2.60$4.10
3$6.00$2.70$4.20
5$10.00$2.90$4.40
10$20.00$3.40$4.90

Assumption: the same 1M-token prefix, reused within the selected lifetime. Output excluded. Cached = one write + (uses − one) reads; fresh = uses × input rate.

Sources: this page; published model prices (data (JSON)); task assumptions (data (JSON)). Published prices · Observed responses · Calculated scenarios · Assumptions stated with each chart.