AI Economics
What will this AI feature actually cost?
Most AI budgets are wrong because the prompt is longer than anyone remembers — system instructions, few-shot examples and retrieved context all bill on every single call. Paste your real prompt and compare eight models on the same workload.
tokens out per call (~750 words = 1,000 tokens)
30,000 calls per month
- Token counts are estimated, not tokenised — typically within about 10% on ordinary prose, less accurate on code and non-Latin scripts. Good enough to plan a budget, not to reconcile an invoice.
- Prices change frequently. This table was last reviewed in mid-2026; confirm against the provider before you commit to anything.
- INR figures use an indicative rate of ₹88 to the dollar and exclude GST.
- Excluded: caching discounts, batch pricing, fine-tuning, and rate-limit tiers — all of which can move a real bill significantly in either direction.
- Everything runs in your browser. No prompt is transmitted or stored.
Building something where these numbers matter?
We design AI systems around real unit economics — model routing, caching and prompt budgets that hold up at production volume.