Monthly workload
100M input tokens and 10M output tokens.
OpenAI use-case brief
The cheapest OpenAI model for extraction is not always the cheapest row.
This page separates the absolute cheapest token row from the cheapest practical extraction default, then shows when long-document extraction pushes the answer back upmarket.
Scenario map
Different extraction workloads have different cheapest answers.
The right answer changes once you separate small-turn extraction, batch extraction, and long-document extraction instead of treating them as the same job.
| Scenario | Cheapest token row | Cheapest viable default | Decision read | Sources |
|---|---|---|---|---|
| Small single-request extraction | gpt-5-nano | gpt-5-mini | gpt-5-nano is the cheapest pure token row, but GPT-5 mini is often the safer cheap default when output quality or structured extraction consistency matters more than absolute minimum cost. | |
| High-volume batch extraction | gpt-5-nano batch | gpt-5-mini batch | At pure volume, gpt-5-nano batch is the cheapest headline row. GPT-5 mini batch becomes the safer practical default when the workflow cannot tolerate the cheapest possible quality floor. | |
| Long-document extraction | gpt-5-mini on paper | gpt-5.4 when context is the real constraint | Once the extraction path needs very large prompt or retrieval context, GPT-5.4 can become the cheapest row that still fits, even though it is not the cheapest token row. |
Worked example
Price the extraction job before assuming the cheapest headline row is good enough.
This example compares three candidate rows under a high-volume structured extraction workload so the cheap-row versus viable-row distinction is visible.
High-volume structured extraction
This sample uses 100M input tokens and 10M output tokens on a repeatable extraction workload with no hosted tools.
Worked example
This compare isolates model pricing so the cheapest-row versus viable-row decision can be seen before hosted tools enter the path.
Shape of work
Repeatable structured extraction with no web search, file search, or runtime.
Compared options
gpt-5-nano batch, gpt-5-mini batch, and gpt-5.4 short batch.
Decision scope
Token pricing only. No hosted tools in this sample.
Model option
gpt-5-nano batch
~$5 per month
Input spend
100M x $0.03 = $3.
Output spend
10M x $0.20 = $2.
Decision read
This is the cheapest headline extraction row in the current pricing table.
Recommended next check
Only choose it if the quality floor is genuinely acceptable for the extraction task.
Model option
gpt-5-mini batch
~$23 per month
Input spend
100M x $0.13 = $13.
Output spend
10M x $1.00 = $10.
Decision read
This is often the cheapest practical default once teams want stronger extraction reliability than the absolute floor row.
Recommended next check
Confirm whether GPT-5 mini quality is good enough before paying up to GPT-5.4.
Model option
gpt-5.4 short batch
~$200 per month
Input spend
100M x $1.25 = $125.
Output spend
10M x $7.50 = $75.
Decision read
This is not the cheap extraction default, but it stays relevant when extraction quality, long context, or a broader tool surface are the real constraints.
Recommended next check
Use this path only if the extraction job genuinely needs the flagship fit instead of just better prompt tuning.
Cheapest by token row
gpt-5-nano batch is the cheapest headline extraction row in this sample.
Cheapest viable default
GPT-5 mini batch is usually the more practical cheap default when teams want a low row without collapsing to the absolute floor option.
When the answer changes
Long-document or broader-tool extraction can move the decision back toward GPT-5.4 even though it is far more expensive on token price alone.
This example deliberately excludes hosted tools so the model-only extraction decision stays visible first.
Recommendation summary
Use one answer for the cheapest row and another for the cheapest workable row.
These cards close the extraction decision without pretending that every extraction workload wants the same model.
Official sources
Check the OpenAI pages behind this recommendation.
This page stays useful only if the source set remains narrow and auditable.
Pricing
Open official pageOpenAI API pricing
Source of record for gpt-5-nano, gpt-5-mini, and gpt-5.4 token pricing and batch rows.
Model page
Open official pageGPT-5 mini model page
Use this when deciding whether GPT-5 mini is the cheapest row that still fits the extraction workload.
Model page
Open official pageGPT-5.4 model page
Use this when long-document or tool-heavy extraction pushes the decision beyond the cheapest mini row.
Continue the site
Keep moving through the decision from here.
Use the groups below to move laterally through the decision, not back out into another doc hunt.
Related pages
Stay in the same decision neighborhood instead of backing out to search.
Comparisons
Side-by-side model comparisons and scenario recommendation pages for cost-sensitive decisions.
Open pageGPT-5.4 vs GPT-5 mini
Side-by-side comparison of GPT-5.4 and GPT-5 mini across price, fit, and tool pressure.
Open pageGPT-5 mini pricing
Single-model pricing brief for GPT-5 mini across standard and batch rows.
Open pageCompare pages
Open the pages that turn this topic into a side-by-side decision.
Replacement pages
Use the likely substitutes, migration targets, or fallback choices as the next click.
Source category pages
Trace the source families behind this page instead of opening random docs in isolation.