CalcSays
AI cost calculators that show the real number — not the naive estimate. Built for engineers and founders paying for AI APIs: every tool models the mechanism naive math misses, with tested engines and dated prices.
Start here
AI Cost Calculator →Price your whole AI product in one place: base conversation plus memory, RAG, tool schemas, images and hidden reasoning — layered on one request stream so nothing is double-counted, with the biggest lever ranked.LLM Token Calculator →Paste any prompt or document and get an honest token estimate — computed in your browser from the providers' own rules of thumb, shown as a range, then carried into any cost calculator in one click.
Chat & text APIs
“What will my LLM calls really cost per month?”
LLM API Cost Calculator →What LLM API calls really cost across GPT, Claude, Gemini & DeepSeek: the effective price per million tokens after output skew and prompt caching — not just the advertised input rate. Live prices, 315 models.Compare LLM API Costs →Compare any two LLM APIs at your real workload: monthly bills with output skew and prompt caching, savings percentage, and the flip point where the answer reverses. 315 models, live prices.Chatbot Conversation Cost →A chatbot re-sends the whole transcript every turn, so cost grows with the square of the conversation length — not turns × one message. Computes the real cost and what prompt caching claws back.Reasoning Token Cost →Reasoning models burn hidden thinking tokens you never see but pay for at the output rate. This computes the real cost from your own reasoning ratio — often several times the visible answer.
Agents & tool use
“Why is my agent so much more expensive than tokens × price?”
AI Agent Cost Calculator →Estimate the real monthly cost of an AI agent across GPT-5.5, Claude, Gemini & DeepSeek. Accounts for quadratic context accumulation, prompt caching, retries, and framework overhead that naive calculators miss.Multi-Agent vs Single-Agent Cost →Should you split a task across multiple agents? Computes the exact step-count break-even: the shared-context duplication tax splitting costs vs the quadratic context growth it saves.Tool Schema Cost Calculator →Every API call re-sends all your tool/function schemas as input tokens. This calculator shows how much of your bill is tool definitions — and how far prompt caching cuts it.
RAG & embeddings
“What does answering from my own documents cost?”
Embedding Cost Calculator →What embedding a corpus really costs: the one-time number, plus recurring churn and query costs, plus what a model migration costs (re-embed everything). Hand-verified prices across OpenAI, Google, Voyage & Cohere.RAG Cost Calculator →What a RAG pipeline really costs: generation over retrieved chunks vs near-free embeddings, and whether RAG even beats stuffing the whole corpus into a cached long context. Live prices.
Cut the bill
“Which lever actually saves money on the stack I already run?”
Prompt Caching Savings Calculator →How much prompt caching really saves across GPT, Claude, Gemini & DeepSeek: models the cache-read discount AND the write premium against your hit rate, computes your break-even, and ranks 315 models by savings.Batch API Savings Calculator →How much the batch API really saves: the 50% discount applies only to traffic that can wait 24h, so your overall saving is waitable-share × 50%. Model it across GPT, Claude & Gemini from live prices.Model Routing Cost Calculator →Routing easy queries to a cheap model saves money — until misrouted queries re-run on the expensive one and pay twice. This computes the blended savings AND the break-even re-run rate.Gemini Context Price Cliff →Gemini Pro re-prices the whole request at a higher rate once your prompt crosses 200K tokens — a 201K prompt can cost ~2× a 199K one. This computes the cliff and what trimming under it saves.
Images, audio & video
“What do multimodal inputs and generation really cost?”
Vision Image Token Cost →How many input tokens does an image really cost? Each provider tokenizes images differently — the same picture is 2-4× more tokens on one model than another. Compares OpenAI, Claude & Gemini from official formulas.AI Image Generation Cost →Is this image model a good price? Compares it against the cheapest model in the SAME quality tier (not a mismatched draft-vs-premium guess), plus the exact Midjourney-subscription-vs-API break-even.Audio Transcribe vs Native Cost →Two ways to answer from audio: transcribe then run text, or send it to a native audio model. Native audio tokens are priced so differently per provider that the cheaper path isn't obvious — this computes both.AI Video Generation Cost →Text-to-video is priced per second, but a usable clip costs per-second × duration × the iterations it takes to get one you'll ship. This computes the real number across Sora, Veo & Runway.
Build vs buy
“Fine-tune, self-host, or subscribe — where's the break-even?”
Fine-Tuning Break-Even Calculator →Should you fine-tune or keep few-shot prompting? Computes the exact call-volume break-even: training cost vs. the recurring example-token waste few-shot pays on every call. Together AI pricing.Self-Hosted GPU vs API Break-Even →Should you rent GPUs and self-host, or keep paying per token? Computes the exact monthly request-volume break-even: the GPU's fixed hourly rental vs the API's variable per-request cost. RunPod pricing.AI Subscription vs API Cost →Should you keep an AI subscription or pay per token via the API? Computes the exact monthly request-volume break-even: the subscription's fixed fee vs the API's variable per-request cost.