Comparing LLM API prices is annoying. Every provider lists "$/1M tokens" in a
different place, and that number tells you nothing until you map it to your actual
usage. So I built a tiny tool that does the mapping.
▶️ Live (no signup): https://emaadshamsi.github.io/llm-cost-calculator/
What it does
Type in your workload — input tokens/request, output tokens/request, requests/day,
and cached-input % — and it ranks the estimated monthly cost across the current
frontier models (Claude Opus 4.8 / Sonnet, GPT-5.5 / 5.4 / mini / nano, Gemini 3.1
Pro / 3.5 / 2.5 Flash, Grok 4.20, DeepSeek V4, Llama 4 Scout, Mistral Large, Qwen3.7),
with a relative-cost bar and a raw price table you can sort.
There are presets for common shapes (chatbot, RAG app, high-volume classifier,
long-context agent).
The thing that jumps out
For the default sample workload (2k in / 500 out / 1k req/day), the spread is wild:
- DeepSeek V4 Flash — free
- Llama 4 Scout — ~$9/mo
- Gemini 2.5 Flash-Lite — ~$12/mo
- …
- Claude Opus 4.8 — ~$675/mo
- GPT-5.5 — ~$750/mo
Same workload, ~80× cost difference. The lesson isn't "always pick the cheap one"
— it's that for high-volume, simple calls you're often lighting money on fire using a
flagship, and a flash/mini tier does the job. Match the model to the task, not the hype.
How it's built
Single index.html, vanilla JS, no dependencies, no backend, no analytics. Prices
live in one array (approximate, as of May 2026 via OpenRouter — always verify live,
provider prices move). Cached input is billed at a rough ~10%.
Code: https://github.com/emaadshamsi/llm-cost-calculator
PRs welcome to keep the prices current. What model/price would you add?
Top comments (0)