Plan your AI budget
Multi-model LLM API cost calculator
Know what your AI will cost, before you build.
Compare 24 models from OpenAI, Claude, Gemini, DeepSeek, Grok and Mistral. Turn your expected usage into a clear monthly API estimate.
02 / Your estimate
OpenAI — GPT-5.4
—
USD per month — token subtotal, excludes cache writes/storage
- Input cost
- —
- Output cost
- —
- Cost per request
- —
- Annual estimate
- —
Need AI integration?
Build an AI workflow that fits your business and your budget.
Hire Dawood →Explore AI automation services03 / Compare your options
Prices verified: 24 models.
Sorted by monthly cost. Price alone does not measure quality or suitability. Custom rates affect only your selected estimate. All quotes are standard paid text rates for input up to 100K tokens per call. The same text can tokenize differently across models. Cache-read savings apply only where a rate is verified.
| Provider / model | Input / 1M | Cached / 1M | Output / 1M | Monthly subtotal | Prices verified |
|---|---|---|---|---|---|
| OpenAI — GPT-5.4 → | $2.5 | $0.25 | $15 | — | 2026-10-08 |
| OpenAI — GPT-6.1 Sol → | $2 | $0.1 | $10 | — | 2026-10-08 |
| OpenAI — GPT-6 Luna → | $0.1 | $0.01 | $0.5 | — | 2026-10-08 |
| OpenAI — GPT-6 Astra → | $10 | $1 | $50 | — | 2026-10-08 |
| Claude — Sonnet 5.5 → | $2 | $0.1 | $10 | — | 2026-10-08 |
| Claude — Opus 5.5 → | $4 | $0.2 | $20 | — | 2026-10-08 |
| Claude — Haiku 5.5 → | $0.1 | $0.01 | $0.5 | — | 2026-10-08 |
| Claude — Haiku 4.5 → | $1 | $0.1 | $5 | — | 2026-10-08 |
| Claude — Sonnet 4.6 → | $3 | $0.3 | $15 | — | 2026-10-08 |
| Claude — Fable 5.1 → | $10 | $0.25 | $50 | — | 2026-10-08 |
| Gemini — 2.5 Flash-Lite → | $0.1 | $0.01 | $0.4 | — | 2026-10-08 |
| Gemini — 3.5 Flash → | $1.5 | $0.15 | $9 | — | 2026-10-08 |
| Gemini — 3.5 Flash-Lite → | $0.3 | $0.03 | $2.5 | — | 2026-10-08 |
| Gemini — 3.1 Flash-Lite → | $0.25 | $0.025 | $1.5 | — | 2026-10-08 |
| DeepSeek — V4.1 Flash (peak) → | $0.3 | $0.006 | $1.2 | — | 2026-10-08 |
| DeepSeek — V4 Pro (peak) → | $1.32 | $0.044 | $3.96 | — | 2026-10-08 |
| Grok — 4.7 → | $2 | Unverified | $6 | — | 2026-10-08 |
| Mistral — Small 4 → | $0.15 | $0.015 | $0.6 | — | 2026-10-08 |
| Mistral — Large 3 → | $0.5 | $0.05 | $1.5 | — | 2026-10-08 |
| Mistral — Medium 3.5 → | $1.5 | $0.15 | $7.5 | — | 2026-10-08 |
| Mistral — Ministral 3 14B → | $0.2 | $0.02 | $0.2 | — | 2026-10-08 |
| Mistral — Ministral 3 8B → | $0.15 | $0.015 | $0.15 | — | 2026-10-08 |
| Mistral — Ministral 3 3B → | $0.1 | $0.01 | $0.1 | — | 2026-10-08 |
| Mistral — Codestral → | $0.3 | $0.03 | $0.9 | — | 2026-10-08 |
Rates checked October 8, 2026. Standard paid text pricing, USD. Verification dates are manually maintained and do not refresh at build time. Excludes batch discounts, cache writes/storage, tools, search, media, hosting, taxes and regional premiums. Rates may change; model links lead to official pricing. Estimates run in your browser.
Understanding LLM costs
How are AI API costs calculated?
Monthly cost = requests x ((input tokens x input price) + (output tokens x output price)) / 1,000,000. Annual estimates assume twelve identical months.
How many tokens are in a prompt?
A rough English estimate is four characters per token. Language, code and each model's tokenizer change the count. Use your provider's tokenizer or API usage logs for accurate estimates.
Is this an OpenAI pricing calculator too?
Yes. Select OpenAI to estimate GPT-5.4 costs, or edit prices for another model. This comparison is limited to 100K input tokens per call, using short-context prices. Larger-context sessions need a separate estimate using provider long-context rates.
How can I lower my AI bill?
Shorten repeated context, limit output length, test smaller models and consider caching or batch processing. Dawood can help build and evaluate your AI agents.
Practical guide
What this tool helps you decide
This LLM cost calculator estimates API token spending for 24 models across six providers. It combines input tokens, output tokens and monthly model calls so you can compare an OpenAI pricing scenario with Claude, Gemini, DeepSeek, Grok and Mistral on the same token workload.
Use it to budget an AI chatbot, retrieval-augmented generation assistant or multi-step agent. These are token-cost comparisons, not rankings of model quality or complete infrastructure bills.
How to use the LLM API cost calculator
Choose a model or a starting workload
Select a provider and model, then choose Chatbot, RAG assistant or Agent as a starting point. Presets are editable examples, not measurements of your application.
Enter average tokens per model call
Include the system prompt, conversation history, retrieved documents and tool responses in input. Include billed reasoning in output. The comparison uses short-context rates and accepts up to 100,000 input tokens per call.
Count every monthly model call
For an agent, one customer task can trigger several calls. Multiply task volume by average calls per task before entering requests per month. A RAG pipeline may also incur embedding and retrieval costs outside this estimate.
Model caching and compare the table
Enable cache reads only if your workload has reusable input. Enter the percentage of input tokens served from cache and inspect the cached rate. Compare monthly subtotals and follow each model link to confirm pricing and availability.
Worked example
A GPT-5.4 chatbot using 1,000 input tokens and 500 output tokens on each of 10,000 monthly calls consumes 10 million input and five million output tokens. The example below uses the model rates in the published comparison table.
- Input token cost
- $25.00
- Output token cost
- $75.00
- Monthly uncached token subtotal
- $100.00
- Monthly subtotal at 50% cache-read share
- $88.75
Illustrative scenario, not a forecast or service quote. Change the inputs in the calculator to match your circumstances.
How to interpret your results
The monthly subtotal combines input and output token charges. Cache-read savings affect only the input portion. Provider tokenizers differ, so sending the same text to two models does not necessarily consume the same number of tokens; validate with usage logs before committing a production budget.
Common input mistakes to avoid
- Count tool loops and retries rather than treating every user task as one request.
- Do not interpret a cache-read subtotal as including cold writes or cache storage.
- Compare accuracy, latency and task completion alongside token prices; a cheaper model may need more calls.
Questions before you use the results
Does the calculator include vector databases and embeddings?
No. It estimates the selected generation model token subtotal. Add embeddings, retrieval, vector database hosting, search tools and other infrastructure separately when budgeting a RAG application.
How should I budget an agent workflow?
Measure average tokens and calls across complete tasks, including failed attempts and tool loops. Enter total monthly model calls, then add costs for external tools and infrastructure separately.
Are the verification dates updated automatically?
No. Each catalog entry has a manually maintained verification date and an official source. A site build does not refresh the date. Prices and model availability can change after verification.