Model cost worksheet - 2026-11-03 Record model, endpoint, input tokens (including cached tokens), output tokens, cache reads/writes, tool calls, region, request tier and rate source. Token counts can differ by model. Use the rate for the whole request when a provider's threshold is exceeded. Calculate each attempt, including failures and retries. Standard uncached example with 2,000 output tokens, US dollars before taxes or premiums: Input tokens Haiku 5.5 Luna 6 Sol 6.1 100,000 0.011 0.011 0.220 100,001 0.0550005 0.0110001 0.220002 272,000 0.141 0.0282 0.564 272,001 0.1410005 0.0559002 1.118004 Haiku rates are $0.10/$0.50 per million input/output through 100,000 input, and $0.50/$2.50 above it. Luna rates are $0.10/$0.50 and Sol $2/$10; above 272,000 input, OpenAI input is doubled and output multiplied by 1.5. Verify current terms before budgeting. Decisions standard examples for Luna: 5,000 fresh input = $0.0005; 80,000-token 30-minute cache write + 1,000 fresh = $0.0101; 80,000 cache-hit + 1,000 fresh = $0.0009. Output is free; cache writes and reads are billed. Long-context and regional premiums may apply. Cost per accepted outcome = sum of all attempt charges / number of accepted outcomes. Add tool charges and, separately, review labor if relevant. Use your observed outcomes rather than an assumed acceptance rate. Compare actual token counts and output quality for each workload. Sources: https://platform.claude.com/docs/en/about-claude/pricing https://developers.openai.com/api/docs/pricing https://developers.openai.com/api/docs/guides/decisions#pricing-and-availability