Cost estimation in Candor is intentionally rough. We take the raw JSON-RPC payload length, divide by 4 to approximate tokens, then multiply against configurable rate cards per model.
It's not exact. It can't be, because the actual token count depends on the LLM's tokenizer and we don't have access to that from the proxy layer. But it's accurate enough to catch the difference between a $0.03 session and a $0.30 session.
If you need exact numbers, export your events and run them through tiktoken yourself. We provide the raw payloads for that reason.
@ButtcoinTNB@moonshot FAKE BUTTCOIN. OG BUTTCOIN CHECK @ButtcoinLOL …it’s verified on JUPITER (✅) and live since 1 month..don’t buy fake ones..
Ca
77KhqraBouiU91KRchLPY8DYbVtfsH4nSBgHCd4UAwBY