Cost of Mind
Every chat turn ships you a receipt. This page is the running total. Tokens in, tokens out, fractional cents per turn, cache hits saving real dollars. The site refuses to hide what it costs to run.
Daily cost, last 14 days
Stacked bars showing cumulative cost per UTC day. Haiku (default, cheap) sits on the bottom. Sonnet (engagement tier) on top. The horizontal line is today's $4 cap.
Recent turns, with receipts
The 3 most recent public turns, with the full pill row attached to every one. Click through for the shareable receipt.
What can you actually do right now?
I can: 1. **Answer questions** on topics from my training (cut off April 2024). I'll hedge when uncertain or out of date, and I'll say "I don't know" rather than guess. 2. **Writ...
What is a token?
A token is the smallest unit a language model processes, roughly four characters of English text per token.
What is a token?
A token is the smallest unit a language model processes, roughly four characters of English text per token.
Privacy. Conversations are anonymous. Each visitor has a 1-year `visitor_uuid` cookie; the prompt and response above are stored against that cookie, never against an email or name. To wipe yours from the database (all conversations, messages, api_calls), use the purge link.
By model
| Model | Calls | Tokens | Cost | Avg / call |
|---|---|---|---|---|
| claude-haiku-4-5 | 2 | 5,208 | $0.4200 | $0.2100 |
| claude-haiku-4-5-20251001 | 1 | 365 | $0.0100 | $0.0100 |
Latency, lifetime average
TTFT is what makes the chat feel alive. Total latency is what you actually pay compute for. Streaming hides the gap between the two.
read the glossary entry →Cache hits, lifetime
Cache reads cost ~10% of fresh input. Each cached token billed at fresh rate would cost ~10x what it actually did. Multiply by total cache reads, take 90% as the realized savings.
read the glossary entry →