Token Cost Calculator

Token Cost Calculator

Compare prices across OpenAI, Claude, Gemini, DeepSeek, MiniMax, and Grok. Real tokenizer for OpenAI, accurate estimates for the rest.

Last updated: 2026-10-20

1,000
2,000
400
1,000 × 2,000in + 400out = 2,000,000 input,400,000 output tokens

Estimated cost across 21 models

Google Cheapest

Gemini 2.5 Flash-Lite

$0.360/ month
$4.32 / year
Input
$0.200
Output
$0.160400,000 tokens
Input
$0.10 / 1M2,000,000 tokens
Output
$0.40 / 1M
Context window
1,000,000Last updated: 2026-10-20
OpenAI

GPT-6 luna

$0.400/ month
$4.80 / year
Input
$0.200
Output
$0.200400,000 tokens
Input
$0.10 / 1M2,000,000 tokens
Output
$0.50 / 1M
Context window
1,050,000Last updated: 2026-10-20
MiniMax

MiniMax M3.1 Flash

$0.540/ month
$6.48 / year

Released 2026-10-19; $0.15 / $0.60 / $0.015 (¥1 / ¥4 / ¥0.1). Flat rate — no long-context tier, and unlike M3 it does not accept service_tier=priority.

Input
$0.300
Output
$0.240400,000 tokens
Input
$0.15 / 1M2,000,000 tokens
Output
$0.60 / 1M
Context window
1,000,000Last updated: 2026-10-20
DeepSeek

DeepSeek V4.1-Flash 1017

$1.08/ month
$12.96 / year
Peak hours

Checkpoint DeepSeek-V4.1-Flash-1017 (released 2026-10-17), served under the API model name `deepseek-flash`; open weights on Hugging Face / ModelScope. The 2026-09-10 launch price is unchanged. The launch checkpoint is kept as its own row (deepseek-flash-0910).

Input
$0.600
Output
$0.480400,000 tokens
Input
$0.30 / 1M2,000,000 tokens
Output
$1.20 / 1M
Context window
1,000,000Last updated: 2026-10-20
MiniMax

MiniMax M3

$1.08/ month
$12.96 / year

Released 2026-06-01. service_tier=priority costs 1.5× ($0.45 / $1.80 / $0.09). No separate cache-write charge. The cached-input rate above 512k doubles to $0.12; cost.ts applies the input multiplier to the cached rate as well.

Input
$0.600
Output
$0.480400,000 tokens
Input
$0.30 / 1M2,000,000 tokens
Output
$1.20 / 1M
Context window
1,000,000Last updated: 2026-10-20
Google

Gemini 3.1 Flash-Lite

$1.10/ month
$13.20 / year
Input
$0.500
Output
$0.600400,000 tokens
Input
$0.25 / 1M2,000,000 tokens
Output
$1.50 / 1M
Context window
1,000,000Last updated: 2026-10-20
Google

Gemini 2.5 Flash

$1.60/ month
$19.20 / year
Input
$0.600
Output
$1.00400,000 tokens
Input
$0.30 / 1M2,000,000 tokens
Output
$2.50 / 1M
Context window
1,000,000Last updated: 2026-10-20
Google

Gemini 3.8 Flash

$3.00/ month
$36.00 / year

Introductory pricing: $0.75 / $3.75 / $0.075 through 2026-12-31, then $1.50 / $7.50 / $0.15 from 2027-01-01.

Input
$1.50
Output
$1.50400,000 tokens
Input
$0.75 / 1M2,000,000 tokens
Output
$3.75 / 1M
Context window
1,000,000Last updated: 2026-10-20
Anthropic

Claude Haiku 4.5

$4.00/ month
$48.00 / year
Input
$2.00
Output
$2.00400,000 tokens
Input
$1.00 / 1M2,000,000 tokens
Output
$5.00 / 1M
Context window
200,000Last updated: 2026-10-20
DeepSeek

DeepSeek V4-Pro

$4.22/ month
$50.69 / year
Peak hours
Input
$2.64
Output
$1.58400,000 tokens
Input
$1.32 / 1M2,000,000 tokens
Output
$3.96 / 1M
Context window
1,000,000Last updated: 2026-10-20
Grok

Grok 4.7

$6.40/ month
$76.80 / year
Input
$4.00
Output
$2.40400,000 tokens
Input
$2.00 / 1M2,000,000 tokens
Output
$6.00 / 1M
Context window
500,000Last updated: 2026-10-20
Google

Gemini 2.5 Pro

$6.50/ month
$78.00 / year
Input
$2.50
Output
$4.00400,000 tokens
Input
$1.25 / 1M2,000,000 tokens
Output
$10.00 / 1M
Context window
1,000,000Last updated: 2026-10-20
OpenAI

GPT-6 sol

$8.00/ month
$96.00 / year
Input
$4.00
Output
$4.00400,000 tokens
Input
$2.00 / 1M2,000,000 tokens
Output
$10.00 / 1M
Context window
1,050,000Last updated: 2026-10-20
Anthropic

Sonnet 5

$8.00/ month
$96.00 / year
Input
$4.00
Output
$4.00400,000 tokens
Input
$2.00 / 1M2,000,000 tokens
Output
$10.00 / 1M
Context window
1,000,000Last updated: 2026-10-20
Google

Gemini 4 Argon

$8.00/ month
$96.00 / year

Introductory pricing: $2 / $10 / $0.10 through 2026-11-30, then $4 / $20 / $0.20 from 2026-12-01. Implicit caching only — explicit caching is not supported. Batch and Flex run at half the standard rate; Priority is 1.8×.

Input
$4.00
Output
$4.00400,000 tokens
Input
$2.00 / 1M2,000,000 tokens
Output
$10.00 / 1M
Context window
1,048,576Last updated: 2026-10-20
Google

Gemini 3.1 Pro Preview

$8.80/ month
$106 / year
Input
$4.00
Output
$4.80400,000 tokens
Input
$2.00 / 1M2,000,000 tokens
Output
$12.00 / 1M
Context window
1,000,000Last updated: 2026-10-20
OpenAI

GPT-5.6 sol

$16.00/ month
$192 / year

Promotional pricing ($4 / $20 / $0.40) is available at least through 2026-11-21; the pre-promo rate was $5 / $30. Re-check after that date.

Input
$8.00
Output
$8.00400,000 tokens
Input
$4.00 / 1M2,000,000 tokens
Output
$20.00 / 1M
Context window
1,050,000Last updated: 2026-10-20
Anthropic

Opus 5.5

$16.00/ month
$192 / year
Input
$8.00
Output
$8.00400,000 tokens
Input
$4.00 / 1M2,000,000 tokens
Output
$20.00 / 1M
Context window
1,000,000Last updated: 2026-10-20
OpenAI

GPT-6 astra

$40.00/ month
$480 / year
Input
$20.00
Output
$20.00400,000 tokens
Input
$10.00 / 1M2,000,000 tokens
Output
$50.00 / 1M
Context window
1,050,000Last updated: 2026-10-20
Anthropic

Fable 5.1

$40.00/ month
$480 / year
Input
$20.00
Output
$20.00400,000 tokens
Input
$10.00 / 1M2,000,000 tokens
Output
$50.00 / 1M
Context window
1,000,000Last updated: 2026-10-20
OpenAI

GPT-5.6 cyber

$55.00/ month
$660 / year
Input
$25.00
Output
$30.00400,000 tokens
Input
$12.50 / 1M2,000,000 tokens
Output
$75.00 / 1M
Context window
400,000Last updated: 2026-10-20

Prices verified 2026-10-20. Source: official provider pricing pages. · Machine-readable: /api/prices.json

Quick takeaways
  • 1Output tokens usually cost 4–8× more than input — keeping replies short is the highest-leverage cost saver.
  • 2Anthropic prompt caching drops repeat-input cost by ~90%; OpenAI cached input drops it by ~90% on GPT-5.
  • 3DeepSeek off-peak hours are ~50% cheaper than peak — schedule batch jobs accordingly.
  • 4Gemini 2.5 Pro doubles its input price above 200K context — large-context calls cost roughly twice as much.
Token Cost Calculator

How to bring the bill down

  1. 1Use a smaller model for routing / classification — Haiku 4.5 / GPT-5 nano cost roughly 5–10× less than the flagship.
  2. 2Cache repeat prefixes (system prompts, tool schemas) — Anthropic cached input drops the rate by ~90%.
  3. 3Set max_tokens explicitly — output is the expensive side; capping it prevents runaway replies.
  4. 4Batch non-urgent jobs into DeepSeek off-peak hours for ~50% cheaper inference.
  5. 5Trim long context — Gemini 2.5 Pro doubles its input rate above 200K; chunk documents that don't need the rest.
Token Cost Calculator

Frequently asked questions

How much does 1 million tokens cost?
It depends entirely on the model. GPT-5 nano is around $0.05 input / $0.40 output per 1M tokens; Claude Opus 4.5 is $5 / $25; Gemini 2.5 Flash is $0.30 / $2.50. Use the calculator above to compare every major model at your exact usage.
Why is output more expensive than input?
Every LLM provider charges more per output token than per input token — typically 4× to 8× more. Output tokens are billed higher because generation is the compute-heavy step (autoregressive decoding), while input is a single forward pass.
How do I count tokens?
For OpenAI models, the calculator uses the real tokenizer (o200k_base for GPT-5, cl100k_base for older families). For Anthropic, Google, DeepSeek, MiniMax, and Grok, it uses a CJK-aware character approximation — accurate to within ~10% for English, ~5% for Chinese.
How accurate is this calculator?
Token counts are exact for OpenAI-family models and approximate for everyone else. Prices are sourced from each provider's official pricing page and re-verified regularly, but providers change prices without notice — always check the source link on each model card before relying on a number for billing.
What is prompt caching?
Prompt caching lets you reuse a previously-seen prefix (system prompt, tool definitions, large document) across requests at a fraction of the normal input cost. Anthropic charges ~90% less for cached reads; OpenAI charges ~90% less on GPT-5; DeepSeek charges ~90% less for cache hits. The biggest single cost saver for repeat-prompt workloads.
Does OpenRouter add a markup?
Yes — OpenRouter charges a small markup (typically around 5%) on top of each provider's official API price. For most single-model workloads the overhead is negligible; OpenRouter wins when you need automatic fallback across providers or a single bill for many models.
Is the Claude subscription cheaper than the API?
Up to a point, yes. Claude Pro ($20/mo) and Max ($200/mo) include generous usage caps. Beyond roughly 1M input + 0.5M output tokens per month (Pro) or roughly 5M input + 2.5M output tokens per month (Max), the API starts winning on pure cost. Use the calculator above to find your break-even point.
How often are prices updated?
Prices are re-verified against each provider's official pricing page regularly. The "Updated YYYY-MM-DD" stamp at the top of the page reflects the last verification. A machine-readable version is available at /api/prices.json — same data, designed for AI agents and programmatic access.
More calculators

Pick the right page for your use case

OpenRouter Pricing Calculator

See what your tokens actually cost after the OpenRouter markup — and when going direct to the provider is cheaper.

Open

Claude Token Cost Calculator

Cost per token and monthly estimates for Claude Opus, Sonnet, and Haiku — with the Anthropic API.

Open

DeepSeek API Pricing

V4-Pro and V4.1-Flash per-token rates, off-peak hours, and prompt-cache savings — compared to Claude, GPT, and Gemini.

Open

ChatGPT API Pricing

GPT-6 and GPT-5 per-token rates, Batch API 50% off, and prompt-cache savings — compared to Claude, Gemini, and DeepSeek.

Open