Terra pricing has five moving parts: fresh input, cached input, cache writes, output, and the mode multiplier — plus one trap, the 272K surcharge. This guide walks the estimate in order. Prefer clicking to arithmetic? The free calculator does all five steps at once; worked examples for common jobs are on the use cases page.
monthly ≈ (fresh × $1.00 + cached × $0.05 + writes × $1.25–2.50 + output × $5.00) × mode — then check the 272K share
Open your OpenAI usage report (or last month's logs from your current provider) and pull three numbers for the period: total input tokens, cached input tokens and output tokens. No report yet? Estimate from a sample: run 100 representative requests, average their token counts, and multiply by your expected monthly request volume.
Example: a support bot at 200,000 conversations × (3,000 in / 400 out) = 600M input and 80M output tokens per month.
Terra bills these twenty-fold apart ($1.00 vs $0.05 per million), so the split matters more than the total. Subtract cached input from total input to get fresh input. If your workload repeats a system prompt, a document set or a conversation prefix, most of your input should land in the cached bucket — check the hit rate in your usage report rather than guessing.
Example: 600M total input at a 70% cache hit rate = 180M fresh + 420M cached.
Writing the cache is not free: $1.25 per million tokens at 30-minute retention, $2.50 at 12-hour retention. Estimate how many tokens your system writes per month — roughly, the size of your cached context × how often it is refreshed. Long-lived contexts written once are nearly free; contexts rebuilt every session are a real line item.
Example: 60M write tokens a month at 30-minute retention = $75.
Multiply the running total by your mode: Standard ×1, Batch/Flex ×0.5, Fast ×2. Mixed workloads split here too — price the interactive share at Standard and the asynchronous share at Batch separately, then add.
Example: the support bot is fully interactive, so its $676 subtotal stays ×1. The invoice pipeline from the use cases page runs entirely in Batch and halves.
Finally, ask what share of your requests carry a prompt over 272,000 input tokens. Those requests bill at 2× input and cache rates and 1.5× output — for the whole request, not just the excess. Enter that share in the calculator's surcharge field, or add the uplift by hand: surcharge ≈ (share) × (that traffic's input + cache cost) + 0.5 × (share) × (its output cost). If the share is large, the real fix is architectural — split or trim the mega-prompts.
Example: 5% of RAG traffic crossing 272K adds roughly 5% to that workload's input line — small here, decisive for document-stuffing pipelines.
Compare your result with the five worked budgets on the use cases page. If your per-request cost lands far from the scenario closest to your workload, a split or a multiplier above is probably off — most often the cache hit rate.
Terra Cost Planner Pack — $9.99
25 pre-built workload scenarios with the monthly Terra budget already calculated, a four-model monthly planner, and a usage-CSV estimator with the formulas written out. Download, plug in your numbers, get a budget.
One-time payment. Not affiliated with OpenAI; rates verified November 4, 2026.
In the OpenAI usage dashboard or your API response logs: total input tokens, cached input tokens and output tokens per period. Those three numbers are everything the estimate needs; multiply by the Terra rates and apply your mode multiplier.
Your usage report lists cached input tokens separately from total input. Divide cached by total input to get the hit rate. If you are pre-launch, assume 0% for a safe upper bound, then re-estimate once real traffic shows the true rate.
Forgetting the 272K surcharge and cache writes. A workload that looks cheap at $1/$5 can double on the requests whose prompts cross 272K tokens, and cache writes at $1.25-$2.50 per million add up on systems that rewrite their context every session.
Independent tool, not affiliated with OpenAI. Prices as published by OpenAI and checked on November 4, 2026 — verify before committing a budget.