Prices checked Nov 22, 2026 · 19 models

LLM API Cost Calculator: Compare Token Prices Across 19 Models

Type in how many tokens a request uses and how many requests you send per month, and the calculator shows the monthly bill for each model. This is not a tokenizer: you enter the counts, we price them. Every rate links to the vendor's own pricing page with the date we checked it (Nov 22, 2026). It covers pay-per-token API usage; for ChatGPT, Claude or Gemini subscriptions, use the true cost calculator.

Your usage

One request = one call to the API.
Count usage in

Count in tokens or words. 1 token ≈ 0.75 words (approximate; real counts vary by tokenizer).

Share of input tokens billed at the cached rate. 0 = none.
Examples:
Cheapest Ministral 3 8B $2.25/mo
Most expensive GPT-6.2 Astra $280.00/mo
Spread 124x between cheapest and priciest $277.75/mo

All 19 models, cheapest first

Monthly cost for the scenario above. Prices checked . All rates are US dollars per 1 million tokens.

LLM API prices per 1 million tokens and monthly cost for the entered scenario, cheapest first
Model Input / 1M Output / 1M Cached in / 1M Monthly cost Source
Ministral 3 8B Mistral · batch 50% off

Batch 50% off.

$0.15 $0.15 $0.015 $2.25 Cheapest Mistral pricing page (opens in a new tab) ↗ checked
Claude Haiku 5.5 Anthropic · batch 50% off

Rate shown is for prompts up to 100,000 tokens. Longer prompts: $0.50 input / $2.50 output.

$0.10 $0.50 $0.01 $3.50 Anthropic pricing page (opens in a new tab) ↗ checked
GPT-6 Luna OpenAI · batch 50% off

Batch: $0.05 / $0.25.

$0.10 $0.50 $0.01 $3.50 OpenAI pricing page (opens in a new tab) ↗ checked
Mistral Small 4 Mistral · batch 50% off

Batch 50% off.

$0.15 $0.60 $0.015 $4.50 Mistral pricing page (opens in a new tab) ↗ checked
DeepSeek V4.1 Flash (deepseek-flash) DeepSeek · 1M context

Peak rate. Off-peak is half: $0.15 / $0.60.

$0.30 $1.20 $0.006 $9.00 DeepSeek pricing page (opens in a new tab) ↗ checked
Gemini 3.5 Flash-Lite Google · batch 50% off

Paid tier rate; a free tier with limits also exists. Input price covers text, image, video and audio.

$0.30 $2.50 $0.03 $15.50 Google pricing page (opens in a new tab) ↗ checked
Mistral Large 4 Mistral · batch 50% off

Batch 50% off.

$0.70 $2.10 $0.07 $17.50 Mistral pricing page (opens in a new tab) ↗ checked
Grok 4.3 xAI · 1M context

Requests of 200K+ tokens are billed at $2.50 / $5.00.

$1.25 $2.50 $0.20 $25.00 xAI pricing page (opens in a new tab) ↗ checked
DeepSeek V4 Pro DeepSeek · 1M context

Peak rate. Off-peak is half: $0.66 / $1.98. Peak = 01:00–04:00 and 06:00–10:00 UTC, Mon–Fri.

$1.32 $3.96 $0.044 $33.00 DeepSeek pricing page (opens in a new tab) ↗ checked
GPT-6.1 Terra OpenAI · batch 50% off

Batch: $0.50 / $2.50.

$1.00 $5.00 $0.05 $35.00 OpenAI pricing page (opens in a new tab) ↗ checked
Grok 4.7 xAI · 0.5M context

Requests of 200K+ tokens are billed at $4.00 / $12.00 for all tokens in the request.

$2.00 $6.00 $0.50 $50.00 xAI pricing page (opens in a new tab) ↗ checked
Gemini 4 Hydrogen Google · 1M context · batch 50% off

Paid tier rate; a free tier with limits also exists. Batch 50% off.

$1.50 $7.50 $0.075 $52.50 Google pricing page (opens in a new tab) ↗ checked
Claude Sonnet 5.5 Anthropic · 1M context · batch 50% off

1M-token context at the standard rate. Batch 50% off.

$2.00 $10.00 $0.10 $70.00 Anthropic pricing page (opens in a new tab) ↗ checked
Gemini 4 Carbon Google · batch 50% off

Promo price through Nov 30, 2026. From Dec 1: $4.00 input / $20.00 output / $0.20 cached.

$2.00 $10.00 $0.10 $70.00 Google pricing page (opens in a new tab) ↗ checked
GPT-6.2 Sol OpenAI · batch 50% off

Batch: $1.00 / $5.00.

$2.00 $10.00 $0.10 $70.00 OpenAI pricing page (opens in a new tab) ↗ checked
Gemini 4 Neon Google · 1M context · batch 50% off

Paid tier rate. Batch 50% off.

$3.00 $15.00 $0.15 $105.00 Google pricing page (opens in a new tab) ↗ checked
Claude Opus 5.5 Anthropic · 1M context · batch 50% off

1M-token context at the standard rate. Batch 50% off.

$4.00 $20.00 $0.20 $140.00 Anthropic pricing page (opens in a new tab) ↗ checked
Claude Fable 5.2 Anthropic · batch 50% off

Batch 50% off input and output.

$8.00 $40.00 $0.20 $280.00 Anthropic pricing page (opens in a new tab) ↗ checked
GPT-6.2 Astra OpenAI · batch 50% off

Batch: $4.00 / $20.00.

$8.00 $40.00 $0.40 $280.00 OpenAI pricing page (opens in a new tab) ↗ checked

How the monthly cost is computed:

monthly cost = (input tokens × input price + output tokens × output price) ÷ 1,000,000 × requests

With a cached share, that share of the input tokens is priced at the cached rate instead. With batch on, models with a published batch discount are multiplied by the discount factor. Cached-write surcharges, priority tiers and regional uplifts are not modeled.

Example, Ministral 3 8B: 10M input × $0.15 + 5M output × $0.15 = $2.25

How to read the numbers

All prices are US dollars per 1 million tokens.

  • Tokens vs words. A token is a chunk of text, not a word. In English, 1 token is roughly 0.75 words, so 1,000 words is about 1,333 tokens. Treat that as an estimate; each vendor's tokenizer counts a little differently.
  • Input vs output. You pay one rate for what you send and a higher rate for what the model writes back.
  • Cached input. If the same prompt prefix repeats, vendors bill the repeated part at a much lower cached rate. The table lists it per model.
  • Batch. Where a vendor states a 50% batch discount, the table says so. It applies to jobs that can wait for results.

Our default scenario: 1,000 input + 500 output tokens per request, 10,000 requests per month. That is 10M input and 5M output tokens a month.

How much does 1 million tokens cost on each major LLM API?

Input runs from $0.10 per 1M tokens (GPT-6 Luna, Claude Haiku 5.5) to $8.00 (GPT-6.2 Astra, Claude Fable 5.2). Output runs from $0.15 (Ministral 3 8B) to $40.00 (the same two flagships).

  • Cheapest output: Ministral 3 8B at $0.15. Priciest output: GPT-6.2 Astra and Claude Fable 5.2 at $40.00 each.
  • Gemini 4 Carbon shows $2.00 in / $10.00 out. That is a promo price through Nov 30, 2026. From Dec 1 it is $4.00 / $20.00.
  • Full list with vendor links and dates: price table.

How much are OpenAI API tokens?

OpenAI's four models in our table range from $0.10 input / $0.50 output (GPT-6 Luna) to $8.00 / $40.00 (GPT-6.2 Astra), per 1M tokens, as listed on OpenAI's pricing page on Nov 22, 2026.

  • GPT-6.2 Sol: $2.00 / $10.00. GPT-6.1 Terra: $1.00 / $5.00.
  • Cached input is $0.01 (Luna) to $0.40 (Astra). Batch is 50% off, so Luna drops to $0.05 / $0.25.
  • On the default scenario: Luna $3.50 a month, Terra $35, Sol $70, Astra $280.

What is the cheapest LLM API per million tokens?

On the default scenario, Ministral 3 8B is cheapest at $2.25 a month ($0.15 in, $0.15 out). GPT-6 Luna and Claude Haiku 5.5 follow at $3.50 each.

  • The math for Ministral: 10M input × $0.15 + 5M output × $0.15 = $1.50 + $0.75.
  • DeepSeek V4.1 Flash costs $9.00 at its peak rates ($0.30 / $1.20) and $4.50 off-peak, when DeepSeek halves the price. Peak is 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday. See our note on DeepSeek peak and off-peak pricing.
  • Haiku 5.5 charges $0.50 in / $2.50 out for prompts over 100,000 tokens, so long prompts cost five times more.

How much does Claude API cost per month for my usage?

On the default scenario, Claude Haiku 5.5 costs $3.50 a month, Sonnet 5.5 $70, Opus 5.5 $140 and Fable 5.2 $280. Rates are on Anthropic's pricing page, checked Nov 22, 2026.

  • Sonnet 5.5: $2.00 in / $10.00 out, so 10M × $2.00 + 5M × $10.00 = $20 + $50 = $70.
  • Cached input is $0.10 per 1M on Sonnet 5.5 ($0.01 on Haiku, $0.20 on Opus and Fable). If every input token were a cache hit, the Sonnet bill would be $1 + $50 = $51.
  • Opus 5.5 and Sonnet 5.5 include a 1M-token context at standard pricing. Batch is 50% off on all four.

How do I convert words to tokens?

Multiply words by about 1.33. We use 1 token ≈ 0.75 English words, so 1,000 words is about 1,333 tokens and a 500-word reply is about 667.

  • It is an estimate. Code, non-English text and unusual formatting usually produce more tokens per word.
  • For billing-grade counts, use the token count the API returns with each response and enter that.

How does LLM API pricing work: input vs output vs cached tokens?

Monthly cost = (input tokens × input price + output tokens × output price) ÷ 1,000,000 × requests. Output costs more than input on almost every model in our table: 5× on most OpenAI and Claude models, 3× on Grok 4.7 and DeepSeek V4 Pro, and up to 8.3× on Gemini 3.5 Flash-Lite ($0.30 in, $2.50 out).

  • Ministral 3 8B is the exception at 1× ($0.15 both ways).
  • Cached input is billed at about a tenth of the normal input price on OpenAI and Claude (GPT-6.2 Astra $0.40 vs $8.00).
  • Grok 4.7 and Grok 4.3 raise the price for requests of 200K tokens or more, and the higher rate applies to every token in that request. Grok 4.7 goes from $2.00 / $6.00 to $4.00 / $12.00.

Is the OpenAI API free?

No. The OpenAI rates in our table are pay-per-token, and we recorded no free tier for OpenAI on its pricing page on Nov 22, 2026. Credits or promos may exist on your account, so check the page itself.

  • Google lists a free tier for Gemini 4 Hydrogen and Gemini 3.5 Flash-Lite on its pricing page.
  • The cheapest paid OpenAI option is GPT-6 Luna: $3.50 a month on the default scenario, $1.75 with the batch discount.

Which API prices changed recently?

Three changes we track: DeepSeek's peak/off-peak pricing (August 2026), the DeepSeek V4.1 Flash price cut (September 2026) and v0's token price cut (November 2026). Gemini 4 Carbon has one coming: its promo ends Nov 30, and the price doubles on Dec 1.

Price changes we track

Vendors change token prices often, and not always loudly. We log each change with its date on all price changes we track, including the DeepSeek cuts and the v0 token prices. If a rate here looks different from the vendor page, the date next to it tells you which of us is older.

How we check prices

Prices are each vendor's standard on-demand rates as published on their pricing pages on Nov 22, 2026. We re-check them monthly and update the date on this page each time. Batch, priority, fast-mode and regional rates are not included unless a note says so. Mistral's own FAQ example quotes Large 4 at a different rate from its docs table; we use the docs table. Meta Llama is not listed because Meta has no first-party price page. For how we handle price data across the site, see how we check prices and how we test.