Free tool
LLM cost calculator for agent workloads
Estimate what an agent workload costs per month on different models. Enter your traffic and token mix, set each model’s price per 1M tokens, and compare — with prompt caching taken into account.
Runs entirely in your browser. Nothing you type is sent anywhere.
Workload
Share of input tokens billed at the cached-input price.
Models & prices
USD per 1M tokens
Example prices — edit to match your provider’s current pricing.
Monthly estimate
44K requests / month · 528M input + 35.2M output tokens / month
| Model | Per request | Input | Output | Monthly | Savings vs most expensive |
|---|---|---|---|---|---|
Frontier modelBaseline | $0.098 | $2,904.00 | $1,408.00 | $4,312.00 | — |
Mid-tier model | $0.0196 | $580.80 | $281.60 | $862.40 | $3,449.60 (80%) |
Budget modelCheapest | $0.00294 | $87.12 | $42.24 | $129.36 | $4,182.64 (97%) |
Formula
requests = requests_per_day × days input_cost = input_tokens × ((1 − cache_hit) × input_price + cache_hit × cached_price) ÷ 1,000,000 output_cost = output_tokens × output_price ÷ 1,000,000 monthly = requests × (input_cost + output_cost) savings = monthly(most expensive) − monthly(model)
This is an estimate. Real bills also depend on reasoning tokens, cache-write surcharges, tiered or long-context pricing, retries and taxes.
How AutoJev estimates cost and savings
- Each model in AutoJev carries an input, output and cache price per 1M tokens — the same three numbers you enter above.
- With the “Lower cost” decision preference, a route sends each request to the cheapest candidate model that meets the request’s capability needs (such as tools and vision). Models with unknown prices are never treated as free.
- Per request, AutoJev estimates cost from the request’s context size and the expected output (the median of recent comparable requests). Cached-input pricing is only applied to the model already holding the conversation’s cache.
- Set a “Cost baseline” model on a route and AutoJev estimates savings per request as baseline cost − selected cost (never below zero). The baseline only affects this estimate — not model selection or billing.
Stop paying frontier prices for every request
AutoJev routes each agent request to the right model — cheap when the task is simple, stronger when it matters — and logs estimated cost locally. Free and open source.