CHAT TOKEN COUNTER
Count tokens for a full conversation, not just one prompt.
Chat APIs bill the entire message array — system prompt, conversation history, and framing overhead — on every request. Build your conversation below to see what one request really costs. Counting runs in your browser; messages are never uploaded.
CONVERSATION
Messages in this request.
Content tokens
0
Format overhead
0
Total input tokens
0
Input cost / request
$0.00
Show the serialized prompt the API actually tokenizes
Estimating output and monthly costs too? Use the full analyzer for projections, or compare per-model rates on the LLM pricing table.
FAQ
Chat token FAQ
Why is my chat request billed more tokens than my messages contain?
Chat APIs wrap every message in a structured format with role markers and separators. For OpenAI models this adds about 3 tokens per message plus 3 tokens to prime the assistant reply. A 10-message conversation carries roughly 33 tokens of overhead on top of the visible content.
Do system prompts count as input tokens?
Yes. The system prompt is sent and billed with every single request in a conversation, which is why a long system prompt multiplies cost across high-volume workloads. Counting it here shows its true per-request weight.
How do tokens grow over a long conversation?
Most chat applications resend the conversation history with each turn, so input tokens grow roughly quadratically over a session: turn N pays for all N-1 earlier messages again. Summarizing or truncating history is the standard fix; this counter shows what any given history snapshot costs.
Is the overhead the same for Claude and Gemini?
The exact framing differs by provider and is not always documented. This page applies the OpenAI-documented overhead constants to all models as a reasonable approximation, and labels the result as an estimate for non-OpenAI providers.