iilib Workbench

LLM API Cost Calculator

Turn published token prices into a real workload cost. Calculate one request or a large batch, compare the same workload across models, or start with a budget such as $10 and estimate how many tokens it can buy.

This calculator uses reviewed USD token rates, including dated and peak/off-peak pricing when documented. Consumer subscriptions such as ChatGPT Plus or Claude Pro are separate products and are not API credits unless the provider explicitly says otherwise.

Start with a real API cost question

Open a prefilled estimate, then change tokens, model, date or pricing mode. These are scenarios, not claims that a model is the cheapest.

Per-workloadGemini 4 Argon: 1,000 requests →5,000 input and 1,000 output tokens each, Standard pricing Fixed budgetWhat does $10 buy? →Gemini 4 Argon Batch, with an 80/20 input-output mix Time-dependent ratesDeepSeek peak vs off-peak →Compare the same workload across documented pricing modes

Priced models

Currency

USD

Verification span

Oct 3, 2026 – Oct 16, 2026

Filter modelsPrimary modelCompare withChoose a second model for a direct same-workload comparison, or leave automatic alternatives selected.

Pricing scenario

Billing modeUsage date

Standard published model pricing is used unless a reviewed model-specific billing scenario is available.

Input tokens / requestOutput tokens / requestCached input / requestRequests

Same assumptions, different model

Compare the current estimate

The token mix stays constant. Your primary model uses its selected tier; alternatives use Standard or DeepSeek off-peak rates. Different tiers may have different eligibility, latency and caching rules.

Same model, different processing modes

Compare API pricing tiers and a $10 token budget

Hold the model and token workload constant to compare each published tier. The $10 estimate uses an input/output mix with zero cache hits; workload cost uses your cached-token and request settings above.

$10 budget input share (%)

Processing modes may differ in availability, latency, restrictions and cache eligibility. Comparison shows token pricing, not guaranteed service equivalence.

Source-backed pricing transparency

Published price schedule, not historical price tracking

This timeline contains documented effective periods for the selected model and tier. A future announced rate is not proof of a past price change.

Verified historical provider price-change events are not yet available. We do not invent historical prices or trends from a present-day price table.

What can change the real bill?

Processing tiers, long-context rules, cache eligibility, peak-hour and holiday exceptions, tool calls, media tokens, regional prices and promotional credits can all change actual cost. Open the model page and official pricing source before making a purchasing decision.

API cost is not a consumer subscription

A monthly assistant plan and an API account often have separate billing and different limits. iilib treats those as different decision problems: this Workbench tool estimates API token spend; Subscription Intelligence will track consumer plan value separately.

How the math works

Cost mode uses date- and tier-specific per-million-token rates. Cached input is included within input tokens, not added on top; request count multiplies the result.

Budget mode creates a blended price from your input/output token mix, then divides the budget by that blended per-million-token price. The result is an estimate, not a quota guarantee.