Compare · Spend Calculator

AI Spend Calculator

Price the same agent workload on every major model. We normalize to cost per completed task, so retries, caching and output-heavy work count the way they do on your bill.

Your workload

Start from

Input tokens / attempt Prompt, files and tool results sent to the model. Output tokens / attempt Code, answers and reasoning the model writes.

Share of input served from cache 0% Repeated context (system prompt, repo files) billed at the cached rate.

Attempts per task 1.0 = right first time. Retries and dead runs push this up. Tasks per month Completed tasks, across your team or product.

Model you use today

Providers

Today · — — per month

Lowest cost · — — per month

Retries cost you — per month on your current model

Cost per completed task, by model

Estimated cost of your workload on each model, sorted from lowest to highest monthly cost.
Model $ per 1M tokens in / out	Per task	Per month	vs today

How we calculate

Per attempt: uncached input × input price + cached input × cached price + output × output price. Multiply by attempts per task for the cost per completed task, then by tasks per month.

Not included: cache-write surcharges, batch discounts (often 50%), regional uplifts, priority modes and taxes. Long-context tiers are applied where the provider publishes them.

Why this is an estimate

Each model family uses its own tokenizer, so the same prompt can come out as a different number of tokens, and a stronger model may finish in fewer attempts. This page assumes the same counts everywhere.

Riva measures the real numbers (tokens, cache hits, retries and dead runs) from your agents' sessions, so the comparison uses your data instead of guesses.

See how Riva measures it →