Calculate cost per AI request, workflow, and successful outcome
Model price alone does not tell you what an AI workflow costs. Include token volume, calls per run, and the success rate to estimate the economics of the work that users actually receive.
- Per request
- $0.009
- Per run
- $0.045
- Failed-run cost per 1,000
- $9.90
At 78% success, 220 of 1,000 runs consume cost without a successful outcome. Shared infrastructure, tools, cache discounts, and review cost are not included.
From token price to outcome economics
The calculator prices input and output tokens for one request, multiplies that amount by the requests in a workflow run, then spreads the complete run cost across the share of runs that succeed. Failed runs stay in the calculation because they still consume paid model usage.
Price the request
Apply separate input and output rates to the token volume.
Count the workflow
Include every model request made during one complete run.
Measure success
Allocate the cost of both successful and failed runs to delivered outcomes.
Public rates used by the model picker
These presets use public rates checked on August 15, 2026, including Anthropic's time-limited Sonnet 5 promotion and its standard rate. The editable custom option is more accurate when you have contracted pricing or use caching, batch processing, long context, or regional endpoints.
| Model | Input / 1M | Output / 1M | Source |
|---|---|---|---|
| GPT-5.6 LunaOpenAI | $1 | $6 | Official pricing |
| GPT-5.6 TerraOpenAI | $2.5 | $15 | Official pricing |
| GPT-5.6 SolOpenAI | $5 | $30 | Official pricing |
| Claude Sonnet 5 (standard rate)Anthropic | $3 | $15 | Official pricing |
| Claude Sonnet 5 (promo through Aug 31)Anthropic | $2 | $10 | Official pricing |
| Claude Opus 5Anthropic | $5 | $25 | Official pricing |
| Gemini 3.5 FlashGoogle | $1.5 | $9 | Official pricing |
| Gemini 3.5 Flash-LiteGoogle | $0.3 | $2.5 | Official pricing |
What this calculator includes and leaves out
Included
- input and output token cost
- multiple requests in one workflow
- the effect of failed runs on delivered outcomes
- directional failed-run cost at volume
Not included by default
- cache write, cache read, batch, or long-context adjustments
- tools, search, grounding, storage, and infrastructure
- human review, support, observability, or gateway cost
- taxes, credits, commitments, and provider bill corrections
An estimate is the start, not the accounting record
Use the result to compare workflow designs or spot where success rate dominates unit cost. For finance and margin reporting, preserve provider usage, pricing versions, business dimensions, and allocation rules, then reconcile telemetry to the bill. The full method is covered in our AI cost attribution guide. To compare model and context changes with controlled tasks, review the AI coding agent benchmarks. If you are choosing a workspace plan, compare trAIce pricing against the event, employee, and retention limits you need.