Cost
What it costs to run Clarity, both to ingest articles and to answer questions.
Clarity is inexpensive to run. Both halves of the system (turning articles into memory, and answering questions from it) run on an efficient model. Cost tracks the amount of text processed, regardless of which feature or analyst persona is in play.
Ingesting articles
Reading one article into Clarity (extracting its entities, facts, relationships, and causal chain, then reconciling it into memory) costs about one cent. Because the work is dominated by fixed prompts, cost rises only gently with article length:
Short (a news brief)
~$0.008
~$0.80
~$8
Typical (a full article)
~$0.009
~$0.90
~$9
Long (a report or transcript)
~$0.011
~$1.10
~$11
So a day's feed of a few hundred articles costs a few dollars to ingest. The step that connects an article's causal chain end-to-end adds only about 15% to that. The surcharge buys memory that can explain why something happened, not the what alone.
Answering questions
Answering a question costs a few cents, roughly two to four times the cost of ingesting one article. A question is more expensive than an article because the agent works iteratively: it retrieves evidence, reasons over it, and composes a cited answer, re-reading its working context at each step. More involved questions that require more steps cost a little more.
All four analyst personas (base, stock analyst, portfolio analyst, investment advisor) cost the same to run. They differ in how they shape the answer, not in price.
Other agents
Clarity's chat is one of several agents, and they fall into two cost tiers:
Text agents, cents per run. Chat, the mobile assistant, and the investment-strategy agent all run on the same efficient model, so they cost a few cents per interaction. A full generated report is the most expensive text operation, roughly 5–50 cents, because it researches across many sources and quality-checks every section.
The voice agent runs differently. Real-time voice uses speech-to-text, text-to-speech, and reasoning models priced by the audio-minute, so a voice session costs more, on the order of 5–20 cents, and scales with conversation length rather than tokens. It is the one agent on a separate, pricier model stack.
What drives cost
Text volume. Longer articles and questions that pull in more evidence cost more. The relationship is gentle rather than linear.
One model across features. Ingestion and answering share one model, so there is no premium for a particular persona, surface, or specialization.
Predictable at scale. Ingesting ten thousand typical articles is on the order of ninety dollars. The cost of a Clarity-backed product is dominated by question volume, which you control.
Last updated