Cost Forecaster
Forecast monthly LLM costs for chatbots, agents and batch jobs, with an honest accuracy label on every number.
New here? Start from an example:
Cost Forecast
Cost Breakdown
Sensitivity how each input (±20%) moves your monthly cost
What If drag to watch the monthly cost respond
Cache Warmth more arrivals keep the cache warm, so cost per message falls
Cost vs Context cost rises with accumulated context, then the window truncates it
Share & Export
Share as Image
How the Forecast Works
Chatbots and agents re-send the system prompt and the whole history on every turn, so cost grows faster than the number of turns. The forecaster adds that up per turn, caps it at the model's context window, and prices input and output separately. Prompt batches are a single call repeated at your volume.
Questions
Is anything uploaded?
No. Everything runs in your browser. The share link stores only your settings, never pasted text.
How accurate is it?
It is a model, not a quote. The badge shows whether token counts are exact (OpenAI, when the tokenizer loads) or estimated. Real bills also depend on retries, tool calls and provider rules.
What does prompt caching change here?
With caching on, the repeated part of each turn is billed at a reduced rate. The headline assumes a warm cache, and the Cache warmth chart shows how lower traffic erodes the saving.
Where do prices come from?
From the open-source LiteLLM catalog, refreshed daily, and not verified against provider billing. The footer line at the bottom of the forecast shows the date.