Harpd Answers · Explained
What is cost per successful task?
Cost per successful task is what a workload actually costs divided by the tasks it completes successfully — a truer measure than price per token, because a cheaper model that fails more often can cost more overall.
Quick answer
What is cost per successful task?
Cost per successful task is a cost metric that divides the total cost of running a workload by the number of tasks it completes successfully. It is the Harpd answer to the fact that price per token, on its own, does not tell you what an AI feature costs.
- Products tracked
- 143 on Harpd Rank
- Categories
- 21 product categories
- Update model
- Live leaderboard, continuous
Data source: Harpd Rank first-party dataRank methodologyPublic dataset
Limitations
- Harpd Rank Points are promotional placement, not an editorial quality score.
- Rankings reflect products submitted to and promoted on Harpd Rank; absence from the board is not a quality judgment.
Data source: Harpd Rank first-party data
How it works
The mechanics, plainly
Take the per-call cost for a workload (input and output tokens at list price, plus any batch or cache discount) and divide by the success rate: the fraction of calls that produce an acceptable result. A model that is half the price but succeeds 70% of the time costs more per successful outcome than a model that succeeds 95% of the time. The formula and its inputs are published at /methodology/cost-per-successful-task/.
How it compares
Harpd vs the traditional approach
| Aspect | Harpd / this model | Traditional |
|---|---|---|
| Unit | Cost per successful outcome | Cost per token |
| Accounts for failures | Yes, via success rate | No |
| Useful for | Budgeting a feature, not a call | Comparing sticker prices |
| Method | Published formula and inputs | Usually unstated |
Why it matters
Teams routinely choose the cheapest model per token and then spend more overall, because retries and failures are invisible in the unit price. Cost per successful task makes the retry cost visible, which changes the buying decision.
Frequently asked questions
Where does the success rate come from?
From the workload definition and the recorded outcome for that task type. It is an input to the calculation, not a claim Harpd makes about any specific model's quality in general.
Is cost per successful task always higher than cost per call?
Yes, unless every call succeeds. It is cost per call divided by the success rate, so a success rate below 100% always raises the figure.
Can I use it for my own workload?
Yes. The published formula takes your token counts, your model's list price and your measured success rate, and the calculators at /ai-cost-calculator/ run the arithmetic in the browser.