Tavus developer pricing: how are minutes billed (minimums/rounding), and how do I estimate cost for my use case?

Most developer pricing pages tell you what you’ll pay, but not how your bill actually grows in real time. With Tavus, minutes are the core unit: you’re paying for live, face-to-face AI Humans running at the speed of human interaction. To budget correctly, you need to know how those minutes are counted, what the minimums and rounding rules look like, and how to translate your specific use case into a predictable cost range.

Quick Answer: Tavus developer usage is metered in time spent running real-time AI video agents (AI Humans). You’re billed on actual session duration with platform-defined minimums and rounding increments, then multiplied by the per‑minute rate of your plan. To estimate cost, model your typical session length, concurrency, and monthly volume, then map that onto your plan’s per‑minute pricing.

The Quick Overview

What It Is: A time-based, usage-driven pricing model for Tavus Developer Accounts, where you pay per minute of real-time AI Human interaction (and potentially per ancillary feature, depending on plan).
Who It Is For: Developers, founders, and product teams embedding Tavus’ real-time video agents into apps and workflows, who need to forecast spend across prototypes, pilots, and scale.
Core Problem Solved: Turning fuzzy “AI costs” into a clear, minute-based model you can plug into your unit economics and capacity planning.

How It Works

Tavus Developer pricing is centered around live compute: real-time perception, speech recognition, LLM orchestration, TTS, and Phoenix-4–powered rendering running while an AI Human is active in a session. Think of it as “time with an AI Human in the room.”

At a high level:

Session Time Is Tracked: When your app starts a real-time Tavus session (e.g., a WebRTC call with an AI Human inside your product), Tavus begins tracking time until that session ends or is disconnected.
Minutes Are Rounded and Aggregated: Each session’s raw duration is rounded according to plan rules (minimum session length and rounding increment), then summed across your workspace for the billing period.
Applied to Plan Rates: The total billed minutes are multiplied by your plan’s per‑minute price (and any add-on rates), yielding your usage charges for the period.

Note: Exact per‑minute rates, minimums, and increments depend on your current developer plan and may change over time. Always verify in your Tavus dashboard or with Sales for absolute numbers; use the logic below to reason about your own cost estimates.

Phases of Billing Logic

1. Session Start and Active State

When your app connects to a Tavus AI Human in real time, a session starts.

This might be:
- A user clicking “Talk to AI Human” in your app.
- A support agent handing off to an AI Human in a live flow.
- An in-product onboarding guide where an AI Human greets and walks the user through screenshare.
From the moment Tavus allocates resources to support presence (perception, audio, video, language), the clock starts.

Key implication: You pay for active presence, not just turns of dialogue. If the AI Human is on-screen, watching, listening, and ready, that time is billable.

2. Minimums and Rounding

Every usage-based system has to decide how to handle partial minutes and very short sessions for both performance and billing simplicity.

While exact thresholds can differ by plan, two levers typically apply:

Session minimum: A minimum billable length per session (e.g., the first 30–60 seconds). This ensures that extremely short connections still cover the cost of spinning up rendering, perception, and connection pipelines.
Rounding increment: How partial minutes are rounded (e.g., to the nearest second, 15 seconds, or full minute). Most developer-friendly systems lean toward:
- Per-second tracking under the hood, then
- Rounding up to the next small increment.

In practice, this often looks like:

A short floor (e.g., sessions shorter than X seconds are treated as X seconds).
Beyond that, usage is counted in fine-grained increments (e.g., second-level metering rolled up to minutes at billing time).

How to think about it without exact numbers:

If your conversations are usually 5–20 minutes, minimums barely matter; rounding noise tends to wash out.
If your flows are very short (e.g., 10–30 second “micro-interactions”), minimums and rounding increments have a bigger impact on effective per‑interaction cost, and you’ll want to design around that.

3. Billing Window Aggregation

Over your billing period (monthly, by default):

Tavus aggregates all billable session time across:
- Environments (dev, staging, prod, depending on your plan),
- Regions,
- Products and features using AI Humans.
Total billable minutes × plan rate = your usage charge component.

Some plans will layer in:

Free tier minutes (e.g., a monthly allowance for experimentation).
Volume discounts at higher tiers or committed-use contracts.
Feature-based pricing (e.g., additional cost for certain enterprise capabilities), but the backbone is still minutes.

Features & Benefits Breakdown

Core Feature	What It Does	Primary Benefit
Time-Based Metering	Measures actual AI Human session time, not just API calls	Aligns cost directly with user conversation volume
Fine-Grained Tracking	Tracks session duration at granular time steps, then applies rounding rules	Predictable billing without paying for large unused chunks
Plan-Tuned Minimums	Applies session minimums and increments tuned per plan and performance profile	Protects performance for real-time video while staying fair

Estimating Cost for Your Use Case

The fastest way to estimate Tavus developer pricing for your scenario is to model it like infrastructure, not SaaS seats.

Walk through this in order:

Step 1: Define Your Conversation Model

Answer three questions:

Average session length (L):
How long is a typical real-time interaction with an AI Human?
- Customer support: 8–15 minutes.
- Sales or demo concierge: 10–20 minutes.
- Onboarding or product walk-through: 5–12 minutes.
- Lightweight check-in or triage: 2–5 minutes.
Sessions per user per month (F):
How often does a typical user talk to an AI Human?
- Heavy usage: daily or multiple times/week.
- Medium: a few times per month.
- Light: only at key lifecycle moments (signup, renewal, etc.).
Monthly active users (U):
How many distinct users will realistically reach an AI Human per month, not just sign up?

Back-of-envelope:
Estimated monthly session minutes = L × F × U.

Step 2: Adjust for Rounding and Minimums

Because minimums and rounding exist, raw time won’t always equal billed time. Adjust with a safety factor:

For sessions ≥ 5 minutes:
Use a modest overhead factor of 1.05–1.15 to account for rounding and occasional early hangups.
For sessions in the 1–5 minute range:
Use 1.1–1.25, depending on whether your interactions skew to the shorter side.
For sub‑minute flows:
Talk to Tavus about your design. Minimums matter a lot here; an architecture tweak (e.g., grouping micro-turns into one session) can significantly reduce effective cost.

Adjusted billed minutes ≈ Estimated minutes × overhead factor.

Step 3: Map to Plan Pricing

Once you know your approximate monthly billed minutes:

Check your Tavus Developer plan for:
- Per‑minute rate.
- Any included free minutes.
- Any volume discounts or committed usage tiers.
Apply the formula:
- Billed minutes = max(0, Adjusted billed minutes − Included free minutes)
- Usage cost ≈ Billed minutes × Per‑minute rate

If you’re on an enterprise or committed-use contract, your effective rate may be lower at scale. Plug in the discounted rate from your agreement instead.

Example Cost Scenarios

These examples are illustrative. Replace the placeholder rate with your actual plan rate from the Tavus dashboard or Sales.

Example 1: Early-Stage Prototype

Use case: Product tours with an AI Human onboarding guide.
Assumptions:
- 300 users/month (U)
- Each user has 2 sessions/month (F = 2)
- Each session is 6 minutes (L = 6)
Raw minutes: 300 × 2 × 6 = 3,600 minutes
Overhead factor: 1.1 (rounding + short drops)
Adjusted billed minutes: 3,600 × 1.1 = 3,960 minutes
Suppose plan rate: $0.XX / minute (use your actual rate)

Estimated usage cost ≈ 3,960 × $0.XX

This gives you the rough magnitude. From here, you compare to your expected revenue or savings per user.

Example 2: Scaling AI Support Agent

Use case: AI Human handling Tier 1 support before human escalation.
Assumptions:
- 5,000 users/month reach support (U)
- Each has 1 AI Human session/month (F = 1)
- Average length: 10 minutes (L = 10)
Raw minutes: 5,000 × 1 × 10 = 50,000 minutes
Overhead factor: 1.07 (conversations are long, rounding overhead is small)
Adjusted billed minutes: 50,000 × 1.07 = 53,500 minutes
Plus potential volume discount at this scale.

Now you can stack this against savings from:

Reduced human handle time.
Faster resolution.
24/7 availability in 30+ languages.

Example 3: High-Frequency Short Interactions

Use case: In-product AI copilot that pops in for quick clarifications.
Assumptions:
- 10,000 users/month (U)
- 6 short sessions/month each (F = 6)
- 2 minutes average per session (L = 2)
Raw minutes: 10,000 × 6 × 2 = 120,000 minutes
Sessions are short, so rounding/minimums matter. Use an overhead factor of 1.2.
Adjusted billed minutes: 120,000 × 1.2 = 144,000 minutes

Here, design choices matter a lot:

Can you keep the same “conversation” open while the user moves through a flow?
Can you encourage longer, more complete interactions instead of many tiny sessions?

These decisions shape your effective cost per user as much as the nominal per‑minute rate.

Ideal Use Cases

Best for real-time, high‑trust experiences: Because Tavus is built for face-to-face presence with lifelike facial behavior and micro-expression tracking, time-based pricing maps neatly to experiences where conversation length is naturally tied to value (sales, support, onboarding).
Best for product teams thinking in unit economics: Because minutes are predictable once you know your workflows, you can treat Tavus usage like any other critical infra line item—run scenarios, set budgets, and adjust UX to hit your targets.

Limitations & Considerations

Exact billing rules differ by plan:
Session minimums, rounding increments, included minutes, and per‑minute rates can change over time or vary across self-serve vs enterprise contracts. Always confirm in your Tavus dashboard or with Sales before locking in financial models.
Very short interactions may be less cost-efficient:
If your design involves many sub‑minute sessions, minimums can inflate effective cost per interaction. Consider:
- Keeping a session open across multiple micro-turns.
- Treating “conversation” as the billing unit in your UX (one guided flow, not 10 separate calls).
- Talking to Tavus about pricing structures for your specific pattern.

Pricing & Plans

Tavus offers two main entry paths:

Developer Account:
Build real-time, human-like AI experiences using Tavus APIs and tools. Best for developers, founders, and teams integrating AI Humans into a product. Pricing is usage-based, anchored on minutes of real-time AI Human interaction, with plan-specific rates, free-tier allowances, and potential volume discounts.
PALs Account:
Personal AI companions for individuals—great if you want to experience Tavus as a user rather than embed it. PALs pricing is separate from developer pricing and is oriented around personal use, not metered developer usage.

For up-to-date numbers:

Sign in to your Developer Account to see current per‑minute rates and included minutes.
Or talk to the Tavus team for enterprise and committed-use pricing if you’re planning large-scale deployment across your organization.

Frequently Asked Questions

How are Tavus developer minutes actually billed—per call, per user, or per minute?

Short Answer: Per minute of active AI Human session time, with plan-specific minimums and rounding rules.

Details:
When your application creates a real-time session with a Tavus AI Human, Tavus tracks how long that AI Human is active—seeing, hearing, and responding—all in real time. That duration (subject to per-session minimums and rounding increments) is converted into billable minutes. Those minutes are summed across all sessions in your workspace over the billing period and multiplied by your plan’s per‑minute rate. You’re not billed per unique user or per API call in isolation; it’s fundamentally time-based.

How do I estimate Tavus developer pricing for my specific use case?

Short Answer: Estimate average session length, frequency per user, and monthly active users; multiply to get minutes; apply a rounding overhead factor; then multiply by your plan’s per‑minute rate.

Details:
Start with your product flows. How long will users actually talk to an AI Human? Plug that into the formula:

Raw minutes = Average session length (L) × Sessions per user per month (F) × Monthly active users (U)
Adjusted billed minutes ≈ Raw minutes × overhead factor (1.05–1.25)
Usage cost ≈ max(0, Adjusted billed minutes − included free minutes) × per‑minute rate

Run a low, medium, and high scenario (e.g., conservative, expected, and upper-bound usage). That gives you a realistic range for monthly spend and lets you test whether your unit economics still work as adoption grows.

Summary

Tavus developer pricing is designed around what actually creates value: live, real-time conversation with AI Humans that see, hear, and respond like people do. You pay primarily for the time those agents are present and active in your product, metered in minutes, with plan-specific minimums and rounding rules that keep billing predictable while maintaining sub‑second, Phoenix-4–powered performance.

If you model your average session length, frequency, and user volume, you can predict your monthly minutes with high confidence. From there, pricing is just arithmetic. Design your flows for natural conversations, not fragmented calls, and you’ll get both better user experience and cleaner cost curves.

Next Step

Want exact numbers for your team and help modeling cost for your product?

Get Started