CodeKitHubWhat Is This Tool?
LLM APIs are billed per token, not per request — and input tokens (your prompt, system message, chat history) are priced differently from output tokens (what the model generates), usually at a 4-8x markup for output. That split makes "how much will this cost" hard to eyeball, especially once you're comparing models across providers with different rate cards.
This calculator takes your input and output token counts, multiplies each by the selected model's per-million-token rate, and adds them together — the same math the vendor's billing system does. Add a monthly request volume and it projects a monthly total, useful for budgeting before you commit to a model in production.
Why Use It?
- 19 models across OpenAI, Anthropic, Google, DeepSeek, Qwen, Moonshot, Zhipu and xAI — budget, mid-tier and flagship options, including the major Chinese model APIs and Grok.
- A side-by-side comparison table ranks every model by cost for your exact token counts, so you don't have to calculate each one by hand.
- Separate input/output cost breakdown, since output tokens are priced several times higher than input tokens on every provider.
- Optional monthly request volume for a quick budget projection, not just a single-request estimate.
- Prices verified directly against vendor documentation, with the verification date shown — not scraped or estimated.
- 100% local — no API key, no account, no data leaves your browser.
How to Use
- Select a model from the dropdown, grouped by provider (OpenAI / Anthropic / Google / DeepSeek / Qwen / Moonshot / Zhipu / xAI).
- Enter your input token count (prompt + system message + any context) and expected output token count.
- Read the cost per request, split into input cost and output cost.
- Optionally enter requests per month to see an estimated monthly cost.
- Scroll down to the comparison table to see every model ranked by cost for the same token counts.
Example
Input
GPT-5.6 Terra, 1,000 input tokens, 500 output tokensOutput
Input cost $0.0025 + output cost $0.0075 = $0.01 per requestAt 10,000 requests/month that's roughly $100/month — small per-request costs add up fast at scale, which is exactly why checking the math before choosing a model matters.
Understanding input vs. output cost
Every model on this list prices input and output tokens separately, and the gap between them is large and consistent across providers — output routinely costs 5-6x more per token than input. That means a chatty, verbose model response costs disproportionately more than a long prompt with a short answer, even if the total token count looks similar.
Practical implication: if you're optimizing API spend, trimming output length (shorter system-prompt instructions like "answer in one paragraph") usually saves more than trimming input, token for token.
Rough monthly cost by usage pattern
These are illustrative examples, not quotes — use the calculator above with your own token counts and model choice for an exact number.
| Usage pattern | Typical tokens/request | Example model | Rough monthly cost (10k requests) |
|---|---|---|---|
| Short chatbot replies | 300 in / 150 out | GPT-5.4 Nano | ≈ $2-3/month |
| Data labeling & classification | 500 in / 50 out | DeepSeek V4 Flash | ≈ $1/month |
| Document summarization | 3,000 in / 400 out | Claude Sonnet 5 | ≈ $100/month |
| Code generation assistant | 1,500 in / 1,000 out | Gemini 3.5 Flash | ≈ $110-115/month |
| Long-form content drafting | 1,000 in / 2,500 out | GPT-5.6 Sol | ≈ $800/month |
Frequently Asked Questions
Why is output priced so much higher than input on every model?
Generating tokens (output) requires a full forward pass through the model for each new token, run sequentially, while processing input tokens (the prompt) can be done in parallel in a single batch. That asymmetric compute cost is why every major provider prices output at roughly 4-8x the input rate.
How do I know how many tokens my prompt will use?
Use CodeKitHub's Token Counter tool — paste your actual prompt or system message and it shows the exact token count using the real tokenizer, rather than a rough character-based guess. Output token counts are harder to predict in advance; a reasonable approach is to check actual usage from a few real API responses and use that as your estimate.
Are these prices guaranteed to be current?
No — AI vendor API pricing changes fairly often, sometimes with little notice. The prices in this tool were verified directly against each vendor's official pricing page on the date shown near the calculator. Before making a budget or purchasing decision, check the current page yourself: OpenAI pricing (developers.openai.com/api/docs/pricing), Anthropic/Claude pricing (platform.claude.com/docs/en/about-claude/pricing), and Google Gemini pricing (ai.google.dev/gemini-api/docs/pricing).
Why isn't [my favorite model] in the list?
This tool covers a representative budget/mid-tier/flagship spread from OpenAI, Anthropic, Google, DeepSeek, Qwen, Moonshot, Zhipu and xAI (19 models total) rather than every model each vendor offers, to keep the list easy to scan and the pricing data easy to keep accurate. If you need a model that isn't listed, check its price per 1M tokens on the vendor's page and do the same input-cost + output-cost math by hand.
Does the Gemini 3.1 Pro Preview price account for long prompts?
No — the price shown is Google's ≤200,000-token tier. Gemini's pricing steps up for prompts longer than that threshold, and this calculator doesn't model that surcharge tier, to keep the tool simple. If your prompts regularly exceed 200k tokens, check Google's pricing page for the exact long-context rate.
Are the DeepSeek, Qwen, Kimi and GLM prices in USD?
Yes — all prices in this tool, including the Chinese models, are the official international API rates in USD, so every model can be compared on the same basis. Some of these vendors also offer CNY-billed consoles for mainland China accounts at different (often lower) rates; check the vendor's own console if you're billing in CNY.