Using fallback FX rate · live rate unavailable
Sent with every call — instructions, RAG chunks, tool definitions.
1 = single-shot request. Above 1, history is resent each turn — input grows every turn.
Assumes your app resends full conversation history on every request, which is common but not universal. Apps using truncation, summarization, or server-side conversation state may use fewer tokens than this estimates.
No custom models yet — add one to include it in the comparison table.
Enable Prompt Caching
Blends each model's real cache-read rate with standard input cost, above that model's minimum cacheable prompt length — an estimated savings figure, not exact provider billing
Share of input tokens assumed to hit the cache, billed at each model's real published cache-read rate. RAG apps typically cache the fixed context/system prompt while fresh user input stays full-price — try a lower ratio to model that. Cache-write costs and real hit-rate variability aren't modeled per-request; this is an estimate.
Use Batch API
Applies each model's real published batch rate where available — models with no batch program are unaffected by this toggle
Safety Buffer
Adds 15% to the estimated cost to account for usage variability — not an actual provider charge, just contingency for retries & error margin
Estimated Reasoning-Token Overhead
Bills hidden "thinking" tokens on reasoning-capable models (GPT-5.6, Claude extended thinking, Gemini thinking mode, DeepSeek V4) as extra output — an estimate, since exact reasoning-token billing isn't published per model
Extra hidden output tokens as a % of your visible output, applied only to models flagged reasoning-capable (badge in the table below). 100% roughly doubles billed output on those models. This is an estimate — actual reasoning-token volume depends on prompt difficulty and isn't controllable or predictable in advance.
Live Summary
Per request
$0.00000
Daily cost
$0.00
Monthly cost (30 days)
$0.00
Model Comparison
Sorted cheapest → most expensive| # | Model | Per Request | Monthly | Relative Cost |
|---|
This calculator estimates model/token charges only. Additional provider charges may apply for tools, search or grounding, image or audio generation, embeddings, storage, retrieval, fine-tuning, or other specialized services — those are not included here.
Model pricing is manually reviewed against official provider pricing pages (see badge in header) and may not reflect live provider pricing. Prices are stored in the MODEL_PRICES array — always confirm current rates on each provider's official pricing page before budgeting.