LLM API Cost Calculator

LLM providers charge per million tokens processed, with distinct prices for Input (prompt) tokens and Output (completion/generation) tokens. Output tokens are typically 3x to 5x more expensive than input tokens due to autoregressive generation latency. In addition, prompt caching discounts (up to 90% off input tokens) dramatically lower costs for system prompts and repeated context.

Leading providers (OpenAI, Anthropic Claude, Google Gemini, DeepSeek) cache prefixes of prompts that remain identical across requests (such as system guidelines, codebase context, or few-shot examples). Prompt cache reads cost between 50% and 90% less than raw input tokens.

For high-volume production, models like Gemini 2.0 Flash ($0.10/M input), DeepSeek V3 ($0.14/M input), and GPT-4o mini ($0.15/M input) offer flagship-level reasoning at less than 1/15th the cost of older generation models.