Tavus: pricing, latency & capabilities

Tavus combines Phoenix avatar rendering with its CVI conversation stack. Its rendering speed and complete conversational response time answer different questions.

Sources reviewed:

When to consider Tavus

Consider Tavus when you want a hosted conversation system with perception, expressive rendering, and turn-taking. Evaluate its conversation behavior with the inputs your users will actually provide.

Pricing by mode

Phoenix 4.5 combines expressive rendering with the CVI conversation stack, perception, and turn-taking.

Starter: $22/mo, 60 minutes, no overage. Builder: $59/mo, 175 minutes, then $0.35/min.

Published specifications and billing bases
DetailTavus
ModelPhoenix 4.5 / CVI
Full avatar usage rate$21.00/hr
Full billing basisBuilder overage
Video-only usage rateNot published
Video billing basisSeparate video rate not verified
Full response timingNot verified · Full conversation timing not established
Video timing<130 ms · Audio-to-video rendering
Output resolutionNot published · No verified specification in the cited sources
FPS35–40 FPS · End-to-end engine throughput on slower machines
Custom motionsExpression controls · Emotion and upper-body motion; signature actions not verified
Moderation controlNot verified · An unmoderated or configurable mode has not been verified
DeploymentHosted API

Usage rates are one component of cost. Use the monthly calculator for supported plan allowances and session assumptions.

Estimate monthly cost

Monthly cost examples

Illustrative totals for the selected monthly plans, using five-minute connected sessions. Each row uses one mode for the entire allowance. Video Only excludes an external voice stack. Taxes, avatar preparation, and unlisted add-ons are excluded.

Selected plan estimates, not invoices or a comparison of every available plan
Provider and mode100 min/month1,000 min/month10,000 min/month
Tavus
Full AvatarBuilder monthly
$59.00$347.75$3,497.75
Tavus
Video OnlyBuilder monthly
Not estimatedA matching video-only tariff is not verified.Not estimatedA matching video-only tariff is not verified.Not estimatedA matching video-only tariff is not verified.

Tavus: 30-second minimum per session; duration rounded to the nearest six seconds. Recording add-ons excluded. Billing source ↗

Plan the integration

Review the CVI API and model configuration, then test interruption handling and the complete response path. Builder overage is the pricing basis here; Starter is a different plan with different limits.

Read the latency correctly

The <130 ms Phoenix-4.5 figure measures audio-to-video rendering. It cannot be compared directly with another provider's complete response clock.

<130 ms is rendering latency, not a full conversational turn. Starter has no overage; the hourly comparison uses Builder at $0.35 per extra minute.

Questions to verify before launch

  • Which model and conversation configuration will the session use?
  • How do the minimum duration and rounding rules affect short sessions?
  • What is the measured speech-to-reply delay in your target region?

Official sources

Griffin preview

Tavus also describes Griffin, a separate full interaction system. Its research page says Griffin is not on the public platform yet, while the Phoenix, Raven, and Sparrow stack remains available. Charlie's homepage demo does not expose an exact deployed model here.

Read Griffin research and availability

Continue comparing

Explore all active models · Read the methodology · Explore integration and demo guides