Latent

Pricing

Pay for the answers you score.

Every answer your model writes, scored. Start free, move up as your traffic grows, or run Latent inside your own environment.

Hosted by Latent

Free

See what Latent flags on your own traffic.

$0

5,000 scored answers a month

Scoring pauses at the limit until the next month

Start free
  • Every answer scored per token
  • The review queue, at the standard threshold
  • At 500 answers, see how many more failures a read calibrated to your traffic catches, and turn it on with Pro. Their text is kept for it, 30 days at most, then hashes only.
  • Judge analyses of flags, 5 a day
  • One project and one person
  • 14 days of history, scores and hashes only
Recommended

Hosted by Latent

Pro

Act on flags and fit Latent to your traffic.

$299a month

100,000 scored answers a month

Then $1.00 per 1,000

Get access
  • Everything in Free, plus
  • The sentence and words behind each flag
  • Policy controls: how much gets flagged and where flagged answers go for review
  • Automatic holds: hold, replace or reroute a flagged answer before your user sees it
  • Alerts in Slack, PagerDuty, email or a webhook
  • Calibration on your own answers, $25 a run
  • A different failure standard on request
  • 30 days of history, with text
  • Two projects
  • Judge analyses of flags, 100 a day, then $0.10 each
  • Up to five people in one organization
  • Email support

Hosted by Latent

Team

For production traffic across several apps.

$1,499a month

1,000,000 scored answers a month

Then $0.60 per 1,000, up to 5,000,000

Get access
  • Everything in Pro, plus
  • Judge analyses of flags, 500 a day, then $0.10 each
  • Unlimited people, with roles and email invites
  • Unlimited projects
  • Weekly calibration, included
  • 90 days of history
  • Audit log of every policy change
  • Priority support
  • Dedicated Slack channel

Add-ons

Add-ons are bought separately and billed in addition to your plan.

Run it in your cloudFor Team plans

Run Latent in your own cloud account, next to your model. It reads your own model as it writes, every token, and can stop a made-up answer mid-stream. It covers one deployment and one monitored model endpoint, under Team's data settings. Prompts, answers and activations stay in your network. Latent's console receives scores, hashes and metadata, plus answer text only if you turn it on, and your account and billing stay there. A second endpoint, a DPA, an SLA, custom failure standards or zero-retention terms come with Enterprise.

$1,500a month, in addition to Team
Talk to us
Weekly calibrationFor Pro plans

Calibration on your own answers every week, run for you automatically. Included in Team.

$99a month, in addition to Pro
Talk to us

Enterprise

Latent inside your model.

Your own model's activations, read per token as it writes, and the option to stop an answer mid-stream. Team can add it with the Run it in your cloud add-on.

Where it runs
Deployed in your cloud or on your hardware, so nothing leaves your network. Also available hosted on a dedicated instance, scored by Latent's reader.
Contract
Annual, with the terms a security and procurement review asks for: zero retention, a DPA, a support SLA, custom failure standards and the calibration service.
Price
Custom, set in the contract

Compare plans

FeatureFreeProTeamEnterprise
Scored answers a month5,000100,0001,000,000Any volume
Above the allowanceScoring pauses until the next month$1.00 per 1,000$0.60 per 1,000, up to 5,000,000Set in the contract
PeopleOneUp to five, in one organizationUnlimited, with roles and email invitesUnlimited
ProjectsOneTwoUnlimitedUnlimited
History14 days, scores and hashes only30 days, with text90 days, with textSet in the contract, including zero retention
CalibrationOne run at 500 answers, served when you upgrade$25 a run, or weekly with the add-onWeekly, includedThe calibration service
Judge analyses of flags5 a day100 a day, then $0.10 each500 a day, then $0.10 eachSet in the contract
Policy controlsThe standard thresholdHow much gets flagged and where it goes for reviewAs Pro, with an audit log of every changeAs Team, with custom failure standards
Automatic holds before deliveryNot includedIncludedIncludedIncluded
Alerts in Slack, PagerDuty, email or a webhookNot includedIncludedIncludedIncluded
SupportDocsEmailPriority, in a dedicated Slack channelA support SLA
DeploymentHosted by LatentHosted by LatentHosted by Latent, or your own cloud with the Run it in your cloud add-onA dedicated instance we host, your cloud, or your hardware

Pricing questions

Anything else, ask us. Get in touch

What happens when I reach my plan's limit?

A scored answer is one finished response from your model that Latent scores, from any model and any way in. With the streaming read on, each sentence Latent reads while an answer streams counts as one more. Free pauses scoring at its included volume until the next month. On Pro and Team, usage billing is on from the upgrade, so scoring continues and the extra answers are billed: $1.00 per 1,000 on Pro, and $0.60 per 1,000 on Team, up to 5,000,000 a month. An admin can set a monthly spend limit or turn usage billing off on the console's Billing page; scoring then pauses at the limit, or at the included volume until the next month. We email you at 80% and 100% of your included volume and of your spend limit, saying which mode you are in. Answers past a limit go unscored, and on every plan your app is never blocked or slowed. Judge analyses of flags have a daily allowance. Free stops at 5 until the next day (UTC). On Pro and Team, with usage billing on, analyses past the allowance are $0.10 each; with it off, they stop until the next day. What a scored answer is

How does calibration work, and what do you keep?

Calibration fits Latent to your traffic: a judge labels a sample of your answers as faithful or made up, and Latent sets its threshold from those labels. A run needs 500 scored answers. On Free, Latent runs one when your first 500 answers arrive and shows how many more failures it catches; the calibrated read is served when you move to Pro. On Pro, each run is $25 for up to 1,000 labeled answers: start one from the console or from calibration@runlatent.ai, or add Weekly calibration for $99 a month and Latent runs one every week. Team includes a run every week. Extra runs on Pro and Team are $25 each, charged when the run starts. A run over more than 1,000 answers is billed as one $25 run for each 1,000 answers or part of 1,000, quoted before it starts. How calibration works

On Free, Pro and Team, prompts and answers are sent to Latent's cloud for scoring and kept under your project's retention. Keep scores and hashes only, or strip identifiers in the SDK's redact hook before they leave your process. Free keeps the text of your first 500 answers for your included first calibration, 30 days at most, then hashes only. If you turn off answer text for a project, Latent has no text to label, so turn it back on before a calibration run. The text is then kept for your plan's window and deleted when it ends. What Latent stores

What does yearly billing mean?

A yearly plan costs ten times the monthly price, so two months are free: $2,990 a year for Pro and $14,990 for Team. The allowance of scored answers stays monthly, and usage billing is billed monthly. Questions about an invoice, a receipt or a refund go to billing@runlatent.ai.