LLM API Cost Calculator
Pick any model, plug in your token usage and daily call volume — get per-call, daily, monthly, and yearly costs instantly. 26 real models including Claude Opus 4.8, Sonnet 4.6, Fable 5, DeepSeek V4 Pro, GLM-5.2, Kimi K2.7. Smart recommendations explain why each alternative is cheaper.
Your Usage
Cost Summary
💡 Smart Recommendations
All Models — Monthly Cost at Your Settings
Sorted cheapest → most expensive. = your current selection. Click any row to switch.
| Rank | Model | Provider | Input/M | Output/M | Monthly |
|---|
Frequently Asked Questions
How accurate is this LLM API cost calculator? ▾
Pricing is sourced from each provider's official pricing page and verified monthly (last update: late June 2026). Models include Claude Opus 4.8, Sonnet 4.6, Fable 5, GPT-5, DeepSeek V4 Pro, GLM-5.2, Kimi K2.7, Gemini 2.5 Pro, and 22 more. Cache read pricing is included where supported. For exact billing, always check your provider's invoice.
Why does the recommendation suggest DeepSeek V4 Pro even when I pick Claude or GPT? ▾
Because DeepSeek V4 Pro is the cheapest frontier-quality model on the market (about 1/30 the price of Claude Opus 4.8, 1/3 the price of Claude Sonnet 4.6). It scores within ~5% of GPT-5 on most reasoning benchmarks. The calculator also shows a same-tier alternative so you can keep your current provider if you prefer — for example, if you need Claude's tool-use stability or GPT's multimodal polish.
How does the "drop a tier" recommendation work? ▾
If you're using a frontier-tier model (Claude Opus, GPT-5, etc.) and the calculator shows a frontier-quality option like DeepSeek V4 Pro that's 80-95% cheaper, you're already on the "drop a tier" play. For workloads where reasoning quality matters but extreme frontier capability is overkill, dropping a tier usually saves 70-90% with <5% quality loss. If you need 100% frontier quality (medical, legal, research), stay where you are.
Does prompt cache pricing actually save money? ▾
Yes — for workloads with repeated prompt prefixes (RAG with stable contexts, customer-service bots, code assistants), cache read rates are typically 90% cheaper than fresh-input rates. DeepSeek V4 Flash has the cheapest cache at $0.0028/1M. For one-off completions, set the cache slider to 0%.