LLM Cost Calculator

It depends on the model. As of 1 October 2026, GPT-6.1 Sol and Claude Sonnet 5.5 cost $2 per million input tokens and $10 per million output tokens, so a request with 1,000 input and 500 output tokens costs about $0.007. Small models are much cheaper: GPT-6 Luna is $0.10 and $0.50 per million tokens, about $0.00035 for the same request. Top-tier models such as GPT-6 Astra and Claude Fable 5.1 cost $10 and $50. Multiply by your monthly request volume, and by the number of model calls per task if you run agents.

There are several proven strategies to reduce LLM API costs. First, enable prompt caching where supported; cached input tokens are billed at a small fraction of the normal rate, which cuts input costs substantially for repeated prefixes such as system prompts. Second, use batch APIs for non-real-time workloads to get roughly 50% discounts. Third, choose the right model for each task: use small models such as GPT-6 Luna, Claude Haiku 4.5 or Gemini 3.5 Flash-Lite for simple tasks and reserve larger models for complex reasoning. Fourth, optimize your prompts to reduce token usage by being concise and removing unnecessary context. Finally, implement response caching at the application level to avoid redundant API calls.

Tokens are the fundamental units LLMs use to process text. Roughly, one token equals about 4 characters or 0.75 words in English. LLM providers charge separately for input tokens (your prompt) and output tokens (the model response), with output tokens typically costing 4 to 8 times more than input tokens. For example, if you send a 500-word prompt (~670 tokens) and receive a 200-word response (~270 tokens), you pay for 670 input tokens plus 270 output tokens at the respective rates for your chosen model.

For an agent, find the cost of one model call (input tokens times the input rate plus output tokens times the output rate) and multiply it by the number of model calls the agent makes per task. For cost per user, divide the monthly total by the number of active users. In production, the reliable way to get both is to record token usage on every call with a user and agent identifier, which is what LLM observability tools do.

Self-hosted open-weight models have no per-token API fees, but you pay for GPU capacity, engineering time for deployment and scaling, and on-call support. Self-hosting usually only pays off at sustained high volume or when data must stay on your own infrastructure. Compare a realistic monthly GPU and staffing cost against the API totals this calculator shows before deciding.

Start by estimating three variables: average tokens per request (input + output), requests per day, and your growth rate. For a chatbot, typical requests use 500-2,000 input tokens and 200-1,000 output tokens. For content generation, expect 500-1,500 input and 1,000-4,000 output tokens. Multiply tokens per request by daily volume and your model pricing to get daily cost, then scale to monthly. Add a 20 to 30% buffer for prompt iterations and unexpected spikes. Use this calculator to compare models side by side; you may find that a cheaper model handles most requests well, with a larger model reserved for complex cases.