Tools
LLM API cost
Enter your input and output token volume, the number of calls and the share of the prompt served from cache: the tool compares cost across models and breaks out fresh input, cache reads, cache writes and output. The built-in table only covers rates that can be stated with confidence, with their date; for any other provider, add your own prices rather than trust a guessed figure.
Everything you send per call: system prompt, tool definitions, conversation history and the user message. This is usually the dominant term, and the one people underestimate.
What the model generates per call. Output costs roughly five times input, so a verbose answer hurts more than a long prompt.
Used for the daily and monthly totals. The monthly figure assumes thirty days.
Portion of your input that is a stable prefix served from the prompt cache. A cache read costs about a tenth of a fresh token; the write costs a quarter more, once per cache lifetime.
Cheapest
$50.40
GPT-4o mini
Most expensive
$1,800
Claude Opus 5
Spread
35.7×
dearest / cheapest
Calls counted
30,000
per month
- GPT-4o miniOpenAI · $0.15/$0.6 per MTok$50.40
- Mistral Small 4Mistral · $0.15/$0.6 per MTok$50.40
- DeepSeek FlashDeepSeek · $0.3/$1.2 per MTok*$101
- Claude Haiku 4.5Anthropic · $1/$5 per MTok$360
- Grok 4.3xAI · $1.25/$2.5 per MTok*$360
- Gemini 3.5 FlashGoogle · $1.5/$9 per MTok$576
- Claude Sonnet 5Anthropic · $2/$10 per MTok$720
- GPT-5.6 TerraOpenAI · $2/$12 per MTok$768
- Claude Opus 5Anthropic · $5/$25 per MTok$1,800
Rates as published on 2026-09-10, first-party API, US dollars per million tokens. Bedrock, Vertex and Foundry bill separately. Check the provider before committing to a budget.
The built-in table only covers rates I can state with confidence. For any other provider, enter its published prices here rather than trust a figure I would have guessed.