Processors are the engines that execute Task Runs. Choice of Processor determines the performance profile and reasoning behavior used. Pricing is determined by which Processor you select, not by the Task Run itself. Any Task Run can be executed on any Processor.

​
Choosing a Processor

Choose a processor based on the complexity of your task. Use Lite or Base for simple enrichments, Core for reliable accuracy on up to 10 output fields, and Pro or Ultra when reasoning depth is critical.
ProcessorDescriptionExample Use Case
LiteHandles basic, one-hop information retrieval Tasks. Lowest compute cost and also typically lower latencyIdentifying parked domains
BaseGreat for basic to moderate‑complexity enrichments (e.g., company firmographics). Executes a simpler retrieval and reasoning to fill 1‑6 columns.Finding an individual’s most recent employment information
CoreThe workhorse processor, best on price-performance. Performs multi‑hop reasoning, cross‑validating facts across multiple sources. Ideal for moderate complexity tasks where trust and scale both matter.Summarizing a startup’s funding information
ProDesigned for longer‑running, high‑complexity tasks. Reliably populates up to 20 output fields, with cross-validation, confidence and excerpts.Analyzing the most relevant information for sales meeting preparation
UltraOur most advanced processor. Runs open‑ended research with extensive retrieval and multi‑stage reasoning, human‑quality analyses on some complex tasks.Analyzing a municipality’s major development goals and how they affect different industries

​
Reference

Each processor varies in performance characteristics and supported features. Use the table below to compare latency, output type, basis features, and more.

ProcessorInput TypeOutput TypeStrengthsp25 Latencyp90 LatencyCost (/1000)Max FieldsBasis Features
LiteText, JSONText or JSONBasic metadata, fallback, low latency3s45s5~2 fieldsCitations, Reasoning
BaseText, JSONText or JSONReliable standard enrichments10s84s10~5 fieldsCitations, Reasoning
CoreText, JSONText or JSONCross-referenced, moderately complex outputs13s160s25~10 fieldsCitations, Reasoning, Confidence, Excerpts
ProText, JSONText or JSONStructured extraction with validation and confidence15s180s100~20 fieldsCitations, Reasoning, Confidence, Excerpts
UltraText, JSONText or JSONOpen-ended, multi-stage reasoning for complex research60s600s300~20 fieldsCitations, Reasoning, Confidence, Excerpts
Cost is measured per 1000 Task Runs in USD. For example, 1 Task Run executed on the Lite processor would cost $0.005.

​
Examples

Processors can be used flexibly depending on the scope and structure of your task. The examples below show how to:

  • Use a single processor (like Lite, Base, Core, Pro, or Ultra) to handle specific types of input and reasoning depth.
  • Chain processors together to combine fast lookups with deeper synthesis.

This structure enables flexibility across a variety of tasks—whether you’re extracting metadata, enriching structured records, or generating analytical reports.

​
Sample Task for each Processor

run_result = client.task_run.execute(
    input="Parallel Web Systems (parallel.ai)",
    output="The founding date of the company in the format MM-YYYY",
    processor="lite"
)

​
Multi-Processor Workflows

You can combine processors in sequence to support more advanced workflows.

Start by retrieving basic information with Base:

base_result = client.task_run.execute(
    input="Pfizer",
    output="Who are the current executive leaders at Pfizer? Include their full name and title. Ensure that you retrieve this information from a reliable source, such as major news outlets or the company website.",
    processor="base"
)

Then use the result as input to Core to generate detailed background information:

run_result = client.task_run.execute(
    input={base_result.output.content},
    output="For the executive provided, find their professional background tenure at their current company, and notable strategic responsibilities.",
    processor="pro"
)

This lets you use a fast processor for initial retrieval, then switch to a more capable one for analysis and context-building.