A credit is a provider’s accounting unit. It might represent a request, a processed page, a result or a duration of work. Comparing two credit balances without comparing their rules is like comparing two quantities with the units removed.
Compare the work and the usable output
Begin with a concrete task: retrieve a defined set of records from specified sources within a freshness window. Decide what makes a record usable, including required fields and duplicate handling. Then count the actual charges required to complete that task.
ScraperAPI’s credit documentation describes costs that vary with the target and options. Apify and Firecrawl publish their own pricing rules. Check current terms directly; an old comparison article is not a reliable invoice calculator.
For record-based tasks, a useful comparison is total actual cost divided by unique records that satisfy the task criteria. If no records qualify, that unit cost is undefined, not zero. Keep subscription commitments, unused allowance and one-off implementation work visible alongside marginal usage cost.
Time-based billing needs a clock definition
Charging by time can be understandable, but “time” must be defined. Media duration measures the audio or video processed. Wall-clock duration measures elapsed time. Resource duration measures occupied compute capacity. Ten workers running concurrently can consume more resources than one worker over the same elapsed interval.
TrueFetch’s design uses a USD balance, not a second public credit currency. Each operation uses one unit: result, file, media_minute or request. Media billing measures successfully processed content duration, not machine runtime. Measure seconds, round up per request to 0.01 minute, and apply no minimum duration:
billableMinutes = ceil(contentSeconds / 60 × 100) / 100
For illustration, 73 seconds becomes 1.22 minutes. At a hypothetical $0.01 per minute, one output language costs $0.0122; two languages cost $0.0244. These rates are examples, not published prices. Internal retries do not increase billable content duration.
Unknown duration requires a maxMediaMinutes bound before execution. Reserve the authorized upper bound, settle eligible usage and release the unused amount. Partial delivery must be authorized in advance; an all-or-nothing task does not charge for incomplete delivery. Verified no-caption and no-speech outcomes are free under the initial policy, with an internal budget limiting unbilled execution.
For result billing, zero delivered records means zero charge. A valid negative answer can be billable for a request-based operation if its contract explicitly says so. Different units need different empty-result rules.
Owning the brand does not remove operating costs
A branded service still pays for compute, proxies, model or transcription calls, storage, bandwidth, failed attempts, monitoring, support and payment processing. Choosing not to display a separate platform fee does not make those costs disappear. They must be covered somewhere in the business model.
For users, the practical questions are whether the bill is predictable, whether useful results justify it and whether failures are explained. For operators, the corresponding questions are whether the unit covers its direct cost and whether total contribution covers ongoing obligations.
TrueFetch includes operating costs in the operation rate, with no separate startup, platform or resource fee. The mechanism is decided; rates and runtime verification remain pending. The preview accepts no payments. Read the billing design and pricing approach for the account defaults and release boundary.