Cheapest AI API Calculator
The same open model costs very different amounts from different providers. Enter your usage and see who is cheapest for it.
Use 22 for weekdays only.
Your prompt, instructions, documents and chat history.
The model's reply.
From your current bill or usage dashboard.
Include any thinking tokens.
Reasoning (thinking) modes bill these as output.
Uses each provider's cached-input price. Providers with no cache price charge full price.
DeepSeek's own API is half price off-peak. Peak is 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday (not on Chinese public holidays).
Change any price you've been quoted, or add your own provider. Leave cached empty if there's no cache price. An edited price replaces any off-peak or batch price.
API bills are in US dollars. Enter today's rate to see monthly costs in another currency.
Every provider of this model
Cheapest first for your usage. Prices are US$ per million tokens. Prices checked on 10 October 2026.
The same work on closed models
Is an open model actually cheaper? Closed-model prices from our LLM API cost calculator, checked 9 October 2026.
How the cheapest provider is found
Same model, same tokens: each provider's price per million decides the bill.
- 1
Tokens a month
30,000 × 1,000
= 30M in
30,000 × 500
= 15M out - 2
Price each provider
DeepInfra
30M × $1.30
= $39.00 in
15M × $2.60
= $39.00 out
= $78.00 - 3
Your blended price
$78.00 ÷ 45M
= $1.733
per million - 4
Rank and save
vs Together AI
$99.00 − $78.00
= $21.00 saved
Read more: the formula, a worked example and every price used
Open-weight models such as DeepSeek, Llama, Qwen, Kimi, GLM, gpt-oss and Mistral can be run by anyone, so many companies sell API access to the same model. They all charge per token, with input (what you send) and output (what the model writes) priced separately, but their prices can differ several times over. Which provider is cheapest depends on your mix: a provider with cheap input and dear output wins for long prompts and short replies, and loses for short prompts and long replies.
The formula
tokens a month = requests a day × days × tokens per request
input cost = input tokens × (uncached share × input price + cached share × cached price) ÷ 1,000,000
output cost = (output + thinking tokens) × output price ÷ 1,000,000
monthly cost = input cost + output cost
DeepSeek off-peak: cost = peak cost × (1 − off-peak share) + off-peak cost × off-peak share
blended price = monthly cost ÷ (all tokens ÷ 1,000,000)
saving = dearest monthly cost − this monthly cost
Worked example
1,000 requests a day for 30 days is 30,000 requests. With 1,000 input and 500 output tokens each, that is 30 million input and 15 million output tokens a month. On DeepSeek V4 Pro (0813):
- DeepInfra ($1.30 / $2.60): 30 × 1.30 + 15 × 2.60 = 39.00 + 39.00 = $78.00 a month, a blended $78.00 ÷ 45 = $1.733 per million tokens.
- DeepSeek's own API at peak price, Atlas Cloud and Together AI ($1.32 / $3.96): 30 × 1.32 + 15 × 3.96 = 39.60 + 59.40 = $99.00.
DeepInfra saves $99.00 − $78.00 = $21.00 a month, 21% less than DeepSeek's own API, and $252 a year.
More examples: DeepSeek V4.1 Flash
For the same usage, every listed provider (DeepSeek's own API at peak price, Atlas Cloud, OpenRouter and Together AI) charges $0.30 input and $1.20 output per million, so they tie:
- 30 × 0.30 + 15 × 1.20 = 9.00 + 18.00 = $27.00 a month with any of them.
- If half the input comes from the prompt cache at $0.006, DeepSeek, OpenRouter and Together cost 15 × 0.30 + 15 × 0.006 + 18.00 = $22.59. Atlas Cloud lists no cache price, so it stays at $27.00.
- All of it in DeepSeek's off-peak hours: 30 × 0.15 + 15 × 0.60 = 4.50 + 9.00 = $13.50.
- OpenRouter's batch variant ($0.112 / $0.336): 30 × 0.112 + 15 × 0.336 = 3.36 + 5.04 = $8.40.
Things to check besides price
- Precision: some providers serve compressed (FP8 or FP4) versions, which are cheaper and faster but can answer slightly differently.
- Context length, speed and rate limits differ between providers serving the same model.
- Routers: OpenRouter sends each request to one of the providers it lists, so the price you pay depends on where it lands.
- Data: check where a provider runs and whether it keeps or trains on your prompts.
Prices used
Prices checked on 10 October 2026. Standard pay-as-you-go prices from each provider's own pricing page, in US dollars per million tokens (input / output / cached input where listed). Promotional prices are left out. Prices change often and some providers price by region or precision, so confirm on each provider's page before you commit.
| Model | Provider | Input / output / cached |
|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek (own API) | $0.3 / $1.2 / $0.006 off-peak $0.15 / $0.6 / $0.003 |
| Atlas Cloud | $0.3 / $1.2 | |
| OpenRouter | $0.3 / $1.2 / $0.006 batch $0.112 / $0.336 / $0.00336 | |
| Together AI | $0.3 / $1.2 / $0.006 | |
| DeepSeek V4 Pro (0813) | DeepSeek (own API) | $1.32 / $3.96 / $0.044 off-peak $0.66 / $1.98 / $0.022 |
| Atlas Cloud | $1.32 / $3.96 | |
| Together AI | $1.32 / $3.96 / $0.13 | |
| DeepInfra | $1.3 / $2.6 / $0.1 | |
| DeepSeek V4 Flash (0731) | Atlas Cloud | $0.44 / $1.32 |
| Together AI | $0.14 / $0.28 / $0.03 | |
| DeepInfra | $0.06 / $0.18 / $0.015 | |
| Kimi K3 | Atlas Cloud | $3 / $15 |
| Together AI | $3 / $15 / $0.27 | |
| DeepInfra | $2.85 / $14.25 / $0.285 | |
| GLM 5.3 Flash | Atlas Cloud | $0.15 / $0.5 |
| Together AI | $0.15 / $0.5 / $0.03 | |
| Qwen3.8 Max | Atlas Cloud | $2 / $6 |
| DeepInfra | $1.65 / $4.951 / $0.206 | |
| Qwen3.7 Max | Together AI | $1.5 / $4.5 / $0.3 |
| DeepInfra | $2.5 / $7.5 / $0.5 | |
| Qwen3.5 397B A17B 397B MoE, 17B active | Together AI | $0.6 / $3.6 / $0.35 |
| DeepInfra | $0.45 / $3 / $0.22 | |
| Qwen3 235B A22B (2507) 235B MoE, 22B active | Together AI | $0.2 / $0.6 |
| DeepInfra | $0.09 / $0.55 | |
| Llama 3.3 70B 70B | Together AI | $1.04 / $1.04 |
| DeepInfra (Turbo version) | $0.1 / $0.32 | |
| Gemma 4 31B 31B | Together AI | $0.39 / $0.97 |
| DeepInfra | $0.2 / $0.4 | |
| DeepSeek V3.2 | DeepInfra | $0.26 / $0.38 / $0.13 |
| GLM-5.3 | Together AI | $1.4 / $4.4 / $0.26 |
| Kimi K2.6 | DeepInfra | $0.75 / $3.5 / $0.15 |
| MiniMax M3 | Together AI | $0.3 / $1.2 / $0.06 |
| gpt-oss-120b | Together AI | $0.15 / $0.6 |
| Qwen3 32B 32B | DeepInfra | $0.08 / $0.28 |
| Mistral Small 3.2 24B 24B | DeepInfra | $0.075 / $0.2 |
Provider pricing pages
- DeepSeek (own API) · Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday (except Chinese public holidays); every other hour is off-peak at half price.
- Atlas Cloud
- OpenRouter · A router: it passes your request on to one of the providers it lists.
- Together AI
- DeepInfra
Closed-model prices (GPT-5.6 Luna, Claude Haiku 5.5, Gemini 3.5 Flash-Lite, Gemini 3.8 Flash, Claude Sonnet 5.5) come from the providers' official pages, checked 9 October 2026. Their cached input is taken as about a tenth of the input price, and their batch APIs as half price.
Common questions
What is the cheapest API for DeepSeek?
It depends on the model and your mix of input and output. In the worked example for DeepSeek V4 Pro, DeepInfra came to $78.00 a month against $99.00 on DeepSeek's own API at peak price. DeepSeek's off-peak hours and OpenRouter's batch variant can cut costs further. Prices change often, so check each provider's page.
Why do providers charge different prices for the same model?
Open-weight models such as DeepSeek, Llama, Qwen and Kimi can be run by anyone, so many companies sell API access to the same model and set their own prices. Some serve compressed FP8 or FP4 versions, which are cheaper and faster but can answer slightly differently.
How do I pick the cheapest provider for my usage?
Enter your requests and your input and output tokens. A provider with cheap input and dear output wins for long prompts and short replies, and loses for short prompts and long replies. The calculator works out each provider's monthly cost for your mix, cheapest first.
What is a blended price per million tokens?
It's your total monthly cost divided by all the tokens you use, in millions. For example, $78.00 for 45 million tokens is a blended $1.733 per million. It lets you compare providers with different input and output prices on one number.
Is OpenRouter cheaper than going direct?
Not always. OpenRouter sends each request to one of the providers it lists, so the price you pay depends on where it lands. For DeepSeek V4.1 Flash, the listed providers including OpenRouter all charged the same standard price, but OpenRouter's batch variant was cheaper.
What should I check besides price?
Precision matters, because some providers serve compressed versions. Context length, speed and rate limits also differ between providers serving the same model. Check where a provider runs, and whether it keeps or trains on your prompts, before you send it real data.
Related calculators
Made by Zulu Calculator, part of AlphaZulu LLC. Formulas are shown in full and rule figures are checked against official sources; see how we build calculators.