Free tool
LLM API pricing, normalized.
Input, output, and cached-input prices per million tokens across the major providers — one table, sortable and filterable. You pay per token in and per token out, and output usually costs several times more than input. Updated 2026-07-09.
| Model | Tier | Context | Input $/M | Output $/M ↑ | Cached in $/M |
|---|---|---|---|---|---|
| Google Gemini 2.5 Flash-Lite | Efficient | 1M | $0.100 | $0.400 | — |
| OpenAI GPT-4o mini | Efficient | 128K | $0.150 | $0.600 | $0.075 |
| DeepSeek DeepSeek-V3 | Balanced | 64K | $0.270 | $1.10 | $0.070 |
| DeepSeek DeepSeek-R1 | Reasoning | 64K | $0.550 | $2.19 | $0.140 |
| Google Gemini 2.5 Flash | Balanced | 1M | $0.300 | $2.50 | — |
| Meta Llama 3.1 405B (hosted) | Flagship | 128K | $3.50 | $3.50 | — |
| OpenAI o3-mini | Reasoning | 200K | $1.10 | $4.40 | $0.550 |
| Anthropic Claude Haiku 4.5 | Efficient | 200K | $1.00 | $5.00 | — |
| Mistral Mistral Large | Flagship | 128K | $2.00 | $6.00 | — |
| OpenAI GPT-4o | Balanced | 128K | $2.50 | $10.00 | $1.25 |
| Google Gemini 2.5 Pro | Flagship | 1M | $1.25 | $10.00 | — |
| Anthropic Claude Sonnet 4.6 | Balanced | 200K | $3.00 | $15.00 | — |
| OpenAI o1 | Reasoning | 200K | $15.00 | $60.00 | $7.50 |
| Anthropic Claude Opus 4.7 | Flagship | 200K | $15.00 | $75.00 | — |
Indicative list prices as of 2026-07-09 — always confirm with the provider before budgeting. Batch, volume, and enterprise rates differ. Anthropic and Google figures mirror the prices our gateway uses.
Turn a rate into a real number.
Plug your volume into the calculator, or point the free audit at your own usage to see what you're actually spending — and what a cheaper tier plus caching would reclaim.
Frequently asked questions
Which LLM is the cheapest?+
For most everyday tasks the cheapest capable options are the small/efficient tiers — Google Gemini Flash-Lite, OpenAI GPT-4o mini, and Anthropic Claude Haiku — priced in cents per million tokens. Open models like DeepSeek and Llama hosted at low rates can be cheaper still. The right answer is the cheapest tier that passes your quality bar for the specific task; the table above sorts by input or output price.
What's the difference between input and output token pricing?+
Providers price input (prompt) tokens and output (generated) tokens separately, and output almost always costs more — commonly 3–5×. That's why the table lists both columns: a workload that reads a lot but writes little is cheaper than one that generates long responses.
What are cached input tokens?+
Many providers discount the repeated front portion of a prompt (a fixed system prompt or document) on subsequent calls — often to around 10% of the normal input rate. The 'Cached in' column shows that discounted rate where a provider offers it. It can dramatically cut cost for apps that resend the same context.
How often is this pricing updated?+
These are indicative list prices as of 2026-07-09. Providers change prices and add models regularly, so always confirm with the provider before budgeting. Anthropic and Google figures mirror the rates the Tokenality gateway uses.
How do I estimate my actual bill from these rates?+
Use the LLM cost calculator: enter your model, requests per day, and average input/output token sizes to get a monthly and annual estimate. For your real numbers, the free AI Spend Audit reads a usage export and computes actual spend and reclaimable waste.