Twelve deterministic calculators for LLM, RAG, eval, observability, self-hosting, agent-loop, and automation-workflow spend — same inputs, same numbers, every time. Provider rates are source-dated and refreshed from official rate cards; verify before you commit.
Pick the decision you are trying to make. Every calculator shows its formula, uses source-dated provider rates, and returns the same numbers for the same inputs.
Size and price a single workload before you commit to a provider.
Decide Which provider and model to run a workload on, and the monthly cost.
You enter Tokens per call, calls per month, provider, model.
You get Monthly and annual cost, the main cost driver, and a cheaper-provider table.
Decide How large a prompt and completion are before you price them.
You enter Your text and a tokenizer family.
You get A low/high token range and byte-level measurements.
Decide Whether a document, prompt, and output reserve fit a model context window.
You enter Document tokens, overhead, output reserve, model window size.
You get Headroom or overflow, and whether you need RAG, summarize, or chunk.
Recurring cost of running LLM systems in production — agents, RAG, evals, and observability.
Decide What each successful agent task costs at scale.
You enter Tasks per month, steps, tool calls, tokens per turn, success rate.
You get Cost per successful task and monthly cost, with failure overhead.
Decide What the automation-platform layer of an agent workflow costs, before LLM tokens.
You enter Agent runs per month, app actions per run, MCP tool calls per run.
You get Monthly platform cost on Zapier, Make, and n8n, with MCP tool-call fan-out accounted separately.
Decide Which iPaaS platform is cheapest for a recurring workflow, and the monthly floor.
You enter Workflow runs per month, billable steps per run, optional Make module count.
You get Monthly cost on Zapier, Make, and n8n public-plan floors, normalized across tasks, credits, and executions.
Decide What an end-to-end RAG system costs to run.
You enter Query volume, corpus size, chunking, embedding, vector store, generator.
You get Monthly cost with a per-layer breakdown and the dominant layer.
Decide What a recurring evaluation run costs.
You enter Samples, candidate models, trials, tokens, optional judge pass.
You get Total and per-run cost, plus a parallel-time estimate.
Decide What LLM tracing and observability will cost each month.
You enter Daily requests, log size, retention, provider.
You get Monthly cost at your volume, with a retention adjustment.
When self-hosting, fine-tuning, or quantizing actually pays off versus the API.
Decide Whether to keep paying per call or self-host on a GPU.
You enter Monthly token volume, API rate, GPU type and utilization.
You get Breakeven volume, monthly cost both ways, and a go/no-go read.
Decide Whether fine-tuning pays back versus staying on prompting.
You enter Training tokens and cost, baseline vs fine-tuned rates, monthly volume.
You get Months to payback and the breakeven volume.
Decide Whether a quantized model fits your hardware, and the quality risk.
You enter Parameter count, source and target precision, KV cache, batch size.
You get VRAM needed, a speedup estimate, and an approximate quality delta.