AI token cost calculator
See what an AI model costs per request and per month. Enter how many tokens go in and come out, and how often you call the API. Prices for 24 models from 8 providers, last checked Oct 17, 2026.
| Model | Input / 1M | Output / 1M | Per request | Per month |
|---|---|---|---|---|
| Claude Haiku 5.5Anthropic | $0.10 | $0.50 | $0.00045 | $13.50 |
| gpt-6-lunaOpenAI | $0.10 | $0.50 | $0.00045 | $13.50 |
| GPT-OSS 120BTogether AI | $0.15 | $0.60 | $0.0006 | $18.00 |
| Mistral Small 4Mistral | $0.15 | $0.60 | $0.0006 | $18.00 |
| gpt-5.4-nanoOpenAI | $0.20 | $1.25 | $0.00103 | $30.75 |
| deepseek-flash (DeepSeek-V4.1-Flash)DeepSeek | $0.30 | $1.20 | $0.0012 | $36.00 |
| gemini-3.5-flash-liteGoogle | $0.30 | $2.50 | $0.00185 | $55.50 |
| Mistral Large 4Mistral | $0.68 | $2.09 | $0.00241 | $72.15 |
| Llama 3.3 70BTogether AI | $1.04 | $1.04 | $0.0026 | $78.00 |
| gemini-3.8-flashGoogle | $0.75 | $3.75 | $0.00337 | $101.25 |
| gpt-5.4-miniOpenAI | $0.75 | $4.50 | $0.00375 | $112.50 |
| grok-4.3xAI | $1.25 | $2.50 | $0.00375 | $112.50 |
| deepseek-v4-pro (DeepSeek-V4-Pro-0813)DeepSeek | $1.32 | $3.96 | $0.00462 | $138.60 |
| Muse Spark 1.3Meta | $1.25 | $4.25 | $0.00462 | $138.75 |
| Mistral Medium 3.5Mistral | $1.50 | $7.50 | $0.00675 | $202.50 |
| grok-4.7xAI | $2 | $6 | $0.007 | $210.00 |
| Claude Sonnet 5.5Anthropic | $2 | $10 | $0.009 | $270.00 |
| gemini-4-argonGoogle | $2 | $10 | $0.009 | $270.00 |
| gpt-6.1-solOpenAI | $2 | $10 | $0.009 | $270.00 |
| gemini-3.1-pro-previewGoogle | $2 | $12 | $0.01 | $300.00 |
| Claude Opus 5.5Anthropic | $4 | $20 | $0.018 | $540.00 |
| Muse WonderMeta | $4 | $20 | $0.018 | $540.00 |
| Claude Fable 5.1Anthropic | $10 | $50 | $0.045 | $1,350.00 |
| gpt-6-astraOpenAI | $10 | $50 | $0.045 | $1,350.00 |
Standard pay-as-you-go API prices in US dollars, before tax. Batch APIs are often about 50% cheaper. Some models charge more for very long prompts or will change price soon; see the notes below.
How the cost is calculated
Cost per request = input tokens × input price ÷ 1,000,000 + output tokens × output price ÷ 1,000,000. Monthly cost = cost per request × requests per day × days. With caching, the cached share of input tokens uses the provider's cache-read price instead.
What is a token?
Models read and write text in tokens: pieces of words. In English, one token is about four characters, or roughly three quarters of a word, so 1,000 words is around 1,300 tokens. Each provider uses its own tokenizer, so the same text can be a different number of tokens on different models.
Why do output tokens cost more?
The model has to generate output one token at a time, which takes far more computing than reading the input. That's why output is usually 3–5× the input price, and why long answers, and reasoning models that "think" before answering, cost more than you might expect.
Would running my own model be cheaper?
At high, steady volume it can be. An open model on a rented GPU costs the same per hour however many tokens you push through it. Compare hourly GPU cloud prices or read our guide to GPU servers for self-hosted AI.
Model price table
| Model | Input / 1M | Cached input / 1M | Output / 1M | Context | Notes |
|---|---|---|---|---|---|
| Claude Fable 5.1Anthropic · source | $10 | $0.25 | $50 | 1,000K | Anthropic's top current model. Full 1M context at the standard price (no long-context surcharge). Cached input = cache read price (0.025x base); 5-minute cache writes cost $12.50. |
| Claude Haiku 5.5Anthropic · source | $0.10 | $0.01 | $0.50 | 1,000K | Price shown is for prompts up to 100,000 tokens. Prompts over 100,000 tokens are billed at $0.50 input / $2.50 output / $0.05 cache read per 1M. |
| Claude Opus 5.5Anthropic · source | $4 | $0.20 | $20 | 1,000K | Full 1M context at the standard price. Cached input = cache read price (0.05x base); 5-minute cache writes cost $5. Uses a newer tokenizer that produces roughly 30% more tokens for the same text than Claude 4.6 and earlier. |
| Claude Sonnet 5.5Anthropic · source | $2 | $0.10 | $10 | 1,000K | Full 1M context at the standard price. Cached input = cache read price (0.05x base); 5-minute cache writes cost $2.50. |
| deepseek-flash (DeepSeek-V4.1-Flash)DeepSeek · source | $0.30 | $0.006 | $1.20 | 1,000K | Peak-hour price shown (cache miss input). Off-peak price is half: $0.15 input / $0.60 output / $0.003 cache hit. Peak = 01:00-04:00 and 06:00-10:00 UTC, Mon-Fri (excluding Chinese public holidays). Supports thinking and non-thinking modes. Context listed as 1M. |
| deepseek-v4-pro (DeepSeek-V4-Pro-0813)DeepSeek · source | $1.32 | $0.044 | $3.96 | 1,000K | Peak-hour price shown (cache miss input). Off-peak price is half: $0.66 input / $1.98 output / $0.022 cache hit. Peak = 01:00-04:00 and 06:00-10:00 UTC, Mon-Fri (excluding Chinese public holidays). Supports thinking and non-thinking modes; no vision. Context listed as 1M. |
| gemini-3.1-pro-previewGoogle · source | $2 | $0.20 | $12 | 1,049K | Gemini 3.1 Pro (preview). Price shown is for prompts up to 200K tokens; prompts over 200K are billed at $4 input / $18 output / $0.40 cached per 1M. Output includes thinking tokens. |
| gemini-3.5-flash-liteGoogle · source | $0.30 | $0.03 | $2.50 | 1,049K | Gemini 3.5 Flash-Lite. Same input price for text, image, video and audio. Output includes thinking tokens. Cache storage billed separately ($1.00 per 1M tokens per hour). |
| gemini-3.8-flashGoogle · source | $0.75 | $0.075 | $3.75 | 1,049K | Gemini 3.8 Flash. Price shown is valid through Dec 31, 2026; from Jan 1, 2027 it doubles to $1.50 input / $7.50 output / $0.15 cached. Output includes thinking tokens. Cache storage billed separately ($0.50 per 1M tokens per hour). |
| gemini-4-argonGoogle · source | $2 | $0.10 | $10 | 1,049K | Gemini 4 Argon, Google's flagship reasoning model. Price shown is valid through Nov 30, 2026; from Dec 1, 2026 it doubles to $4 input / $20 output / $0.20 cached. Output includes thinking tokens. Implicit caching only. Context window is a unified input+output limit. Paid tier only. |
| gpt-5.4-miniOpenAI · source | $0.75 | $0.075 | $4.50 | 400K | Previous-generation (GPT-5.4) mini model, still listed. Max input 272K tokens within the 400K context window. No long-context rate listed. |
| gpt-5.4-nanoOpenAI · source | $0.20 | $0.02 | $1.25 | 400K | Previous-generation (GPT-5.4) nano model, still listed. Max input 272K tokens within the 400K context window. No long-context rate listed. |
| gpt-6.1-solOpenAI · source | $2 | $0.10 | $10 | 1,050K | Mid-tier GPT-6.1 Sol, described as near-Astra performance at lower cost. Price shown is for prompts up to 272K tokens; longer prompts are billed at $4 input / $0.20 cached / $15 output per 1M. |
| gpt-6-astraOpenAI · source | $10 | $1 | $50 | 1,050K | OpenAI flagship (GPT-6 Astra). Price shown is for prompts up to 272K tokens; longer prompts are billed on the whole request at $20 input / $2 cached / $75 output per 1M. |
| gpt-6-lunaOpenAI · source | $0.10 | $0.01 | $0.50 | 1,050K | Smallest and cheapest GPT-6 model (high-volume tasks). Price shown is for prompts up to 272K tokens; longer prompts are billed at $0.20 input / $0.02 cached / $0.75 output per 1M. |
| GPT-OSS 120BTogether AI · source | $0.15 | — | $0.60 | 131K | OpenAI's open-weight gpt-oss-120b served on Together AI serverless inference (MXFP4); the price applies to Together only, other hosts differ. No cached-input price listed. |
| grok-4.3xAI · source | $1.25 | $0.20 | $2.50 | 1,000K | Older, cheaper Grok with 1M context. Price shown is for prompts under 200K tokens; at 200K or more the whole request is billed at $2.50 input / $0.40 cached / $5 output per 1M. US regional endpoint costs 1.1x. |
| grok-4.7xAI · source | $2 | $0.50 | $6 | 500K | xAI flagship. Price shown is for prompts under 200K tokens; at 200K or more the whole request is billed at $4 input / $1 cached / $12 output per 1M. US regional endpoint costs 1.1x. |
| Llama 3.3 70BTogether AI · source | $1.04 | — | $1.04 | 128K | Open-weight Meta Llama model served on Together AI serverless inference; the price applies to Together only, other hosts differ. Context listed by Together as 128K. No cached-input price listed. |
| Mistral Large 4Mistral · source | $0.68 | $0.07 | $2.09 | 1,000K | Public preview (v26.10, released Oct 6, 2026). Launch sale price: Mistral's changelog says 50% off for the first two weeks from Oct 6 (so until about Oct 20, 2026); the list price is $1.36 input / $4.18 output / $0.14 cached per 1M. Mistral calls it open-weight, but the weights aren't downloadable yet (expected by Oct 31). |
| Mistral Medium 3.5Mistral · source | $1.50 | — | $7.50 | 256K | Frontier-class multimodal model for agentic and coding use (v26.04). Context shown on the page as 256k. No cached-input price listed. |
| Mistral Small 4Mistral · source | $0.15 | — | $0.60 | 256K | Hybrid instruct/reasoning/coding model (v26.03), Apache 2.0. Context shown on the page as 256k. No cached-input price listed. |
| Muse Spark 1.3Meta · source | $1.25 | $0.15 | $4.25 | 1,049K | API model ID muse-spark-1.3 (released Sep 2, 2026). Standard-mode price shown: your content isn't used to train Meta's models. Contributor mode (model ID muse-spark-1.3-contributor) costs $0.10 input / $0.002 cached / $0.20 output per 1M in exchange for letting Meta train on your content, with a lower limit of 150 requests a minute and no max reasoning. Same price at any prompt length. Web search grounding costs $2.50 per 1,000 searches on top. |
| Muse WonderMeta · source | $4 | $0.20 | $20 | 1,049K | Meta's flagship (API model ID muse-wonder-1.0, released Oct 12, 2026), still labelled preview. Standard mode only: your content isn't used to train Meta's models, and the cheaper Contributor mode isn't offered for this model. Same price at any prompt length (no long-context premium). Web search grounding costs $2.50 per 1,000 searches on top. The stronger Muse Wonder Full isn't on the API; it comes only with Muse Code subscriptions ($50 or $200 a month). |
Prices last checked Oct 17, 2026 on each provider's own pricing page. Spotted a change? Prices move fast; always confirm on the provider's site before you commit.