AI API cost calculator
Token prices are quoted per million, which is a unit nobody budgets in. Enter your monthly volumes to see the actual bill across all 25 models — and how far the ranking moves when output-heavy work is involved.
List prices checked
million
million
At 100M input and 20M output tokens per month, the cheapest option here is GPT-5 nano at $13 per month. Output is the larger share of that bill, so compare on the output column rather than the headline input rate.
| # | Model | Input cost | Output cost | Monthly | If cached |
|---|---|---|---|---|---|
| 1 | GPT-5 nanoOpenAI · $0.050 in / $0.400 out | $5.00 | $8.00 | $13 | $8.50 |
| 2 | Mistral SmallMistral AI · $0.100 in / $0.300 out | $10 | $6.00 | $16 | — |
| 3 | Gemini 2.5 Flash-LiteGoogle · $0.100 in / $0.400 out | $10 | $8.00 | $18 | $11 |
| 4 | DeepSeek-V4-FlashDeepSeek · $0.140 in / $0.280 out | $14 | $5.60 | $20 | $5.88 |
| 5 | GPT-4o miniOpenAI · $0.150 in / $0.600 out | $15 | $12 | $27 | $20 |
| 6 | Cohere Command RCohere · $0.150 in / $0.600 out | $15 | $12 | $27 | — |
| 7 | Grok 3 minixAI · $0.300 in / $0.500 out | $30 | $10 | $40 | — |
| 8 | CodestralMistral AI · $0.300 in / $0.900 out | $30 | $18 | $48 | — |
| 9 | DeepSeek-V4-ProDeepSeek · $0.435 in / $0.870 out | $44 | $17 | $61 | $18 |
| 10 | GPT-5 miniOpenAI · $0.250 in / $2.00 out | $25 | $40 | $65 | $43 |
| 11 | GPT-4.1 miniOpenAI · $0.400 in / $1.60 out | $40 | $32 | $72 | $42 |
| 12 | Gemini 2.5 FlashGoogle · $0.300 in / $2.50 out | $30 | $50 | $80 | $58 |
| 13 | OpenAI o4-miniOpenAI · $1.10 in / $4.40 out | $110 | $88 | $198 | $116 |
| 14 | Claude Haiku 4.5Anthropic · $1.00 in / $5.00 out | $100 | $100 | $200 | $110 |
| 15 | Mistral LargeMistral AI · $2.00 in / $6.00 out | $200 | $120 | $320 | — |
| 16 | GPT-5OpenAI · $1.25 in / $10.00 out | $125 | $200 | $325 | $213 |
| 17 | Gemini 2.5 ProGoogle · $1.25 in / $10.00 out | $125 | $200 | $325 | $231 |
| 18 | GPT-4.1OpenAI · $2.00 in / $8.00 out | $200 | $160 | $360 | $210 |
| 19 | OpenAI o3OpenAI · $2.00 in / $8.00 out | $200 | $160 | $360 | $210 |
| 20 | GPT-4oOpenAI · $2.50 in / $10.00 out | $250 | $200 | $450 | $325 |
| 21 | Cohere Command ACohere · $2.50 in / $10.00 out | $250 | $200 | $450 | — |
| 22 | Claude Sonnet 5Anthropic · $3.00 in / $15.00 out | $300 | $300 | $600 | $330 |
| 23 | Grok 4xAI · $3.00 in / $15.00 out | $300 | $300 | $600 | $375 |
| 24 | Claude Opus 5Anthropic · $5.00 in / $25.00 out | $500 | $500 | $1,000 | $550 |
| 25 | Claude Fable 5Anthropic · $10.00 in / $50.00 out | $1,000 | $1,000 | $2,000 | $1,100 |
Questions
- How do I estimate my token volumes?
- A rough working figure is 750 words per 1,000 tokens for English prose; code runs denser. Multiply the tokens in a typical request by your monthly request count, and remember that in a multi-turn conversation the entire history is resent as input on every turn — which is why input volume grows faster than people expect.
- Why does output cost so much more than input?
- Output is generated one token at a time and cannot be batched the way input processing can, so it is genuinely more expensive to serve. Typical ratios run 4x to 5x. It matters most for reasoning models and agents, where thinking tokens bill as output and often exceed the visible answer several times over.
- What does the 'if cached' column mean?
- Most providers discount input tokens that repeat a prefix they have already processed — a system prompt, a document, a codebase. The column shows the bill if every input token were served from cache, which is the floor rather than a realistic figure. Real workloads land somewhere between the two columns.
- Are these prices current?
- Each model is stamped with the month its rates were checked against the provider's own pricing page, and the model pages link straight to that page. Model pricing moves faster than any other category on this site, so confirm before you build a budget on it.