# Realtime video API pricing: compare rates and billable timelines

> Compare dated Windflow, Reactor, Decart and fal realtime rates. Estimate billable session time, distinguish pricing units and account for retained idle time.

Source: https://windflow.online/compare/realtime-video-api-pricing · Updated 2026-11-09 · Windflow Labs

**Estimate a realtime video API by multiplying its rate by the seconds its contract actually meters.** The meter might cover a ready session, time after generation starts, active generation, or generated output duration. Two prices ending in “per second” can therefore describe different costs for the same user visit.

This page is maintained by Windflow. The figures below are a **November 9, 2026 source check**, in USD, not a live quote or a performance ranking. They compare documented rates and a hypothetical usage timeline. Models with similar names are not assumed to have identical checkpoints, output quality, resolution or serving behavior.

## Published rate examples

The 60-second column is arithmetic on **60 seconds of the row's own billable unit**. It is not necessarily one minute from clicking Connect, nor one minute of finished output.

| Provider and profile | USD per second | USD for 60 billable seconds | Metering boundary to use |
|---|---|---|---|
| Windflow Helios | $0.0017 | $0.102 | From first generation start until release/end, with documented grace excluded |
| Reactor Helios | $0.0017 | $0.102 | From ready until session termination |
| Windflow LongLive 2.0 | $0.0021 | $0.126 | Windflow billable session time |
| Reactor `longlive-v2` | $0.0017 | $0.102 | Reactor ready session time |
| Windflow LingBot-World 2.0 | $0.0024 | $0.144 | Windflow billable session time |
| Reactor `lingbot-world-2` | $0.0070 | $0.420 | Reactor ready session time |
| Windflow SANA-Streaming | $0.0019 | $0.114 | Windflow billable session time |
| Windflow JoyAI-Video-Edit | $0.0021 | $0.126 | Windflow billable session time |
| Decart Lucy 2.6 realtime, standard | $0.02 | $1.20 | Active generation, per Decart's realtime pricing definition |
| Decart Lucy 2.6 realtime, fast | $0.04 | $2.40 | Active generation in the selected fast tier |
| fal `decart/lucy-2-6/realtime` | $0.02 | $1.20 | Per-second price displayed by that endpoint; confirm its precise session billing window in your account |

Windflow figures match its [public price table](https://windflow.online/pricing) and are indicative private-beta rates. Reactor figures were read from its public [pricing JSON](https://api.reactor.inc/pricing), using the selected model's `rate.amount_per_sec_usd`; its [billing reference](https://docs.reactor.inc/resources/billing) defines the ready-time window. A similarly named model is a starting point for comparison, not proof that the service profiles are equivalent.

Decart's [pricing page](https://docs.platform.decart.ai/getting-started/pricing) lists standard and fast Lucy 2.6 realtime rates and distinguishes active-generation time from the generated-second unit used by video jobs. The fal figure comes from the [Lucy 2.6 realtime playground](https://fal.ai/models/decart/lucy-2-6/realtime). That page displays $0.02 per second, but does not by itself fully specify which handshake, idle, pause or reconnect intervals are billed. Confirm those details before turning it into a whole-visit estimate. Decart-direct pricing should not be used as an assumed billing contract for a fal account.

## A worked session timeline

Suppose a Helios user takes the following path. These timestamps are chosen to explain the arithmetic; they are not measured connection or generation times on either provider.

| Elapsed time | Application state | Seconds in this interval |
|---|---|---|
| 0–20 s | Connecting or waiting for capacity | 20 |
| 20–35 s | Ready; user is preparing the opening prompt, generation has not started | 15 |
| 35–155 s | Generation started; user watches and directs it | 120 |
| 155–185 s | Video is paused, but the owner retains the session | 30 |
| At 185 s | Owner explicitly releases/terminates the session | Meter ends |

Under Windflow's documented policy, the example has **150 billable seconds**: the 120-second generation interval plus the 30-second retained pause. At the checked Helios rate, `150 × $0.0017 = $0.2550` before any billing rounding.

Under Reactor's documented ready-time policy, the same hypothetical state timeline has **165 billable seconds**: 15 seconds ready before generation, 120 seconds generating and 30 seconds of retained pause. At its checked Helios rate, `165 × $0.0017 = $0.2805` before rounding. Neither provider charges the first 20 seconds in its stated connecting/waiting states.

The difference in this example comes from the defined meter start, not from a measured speed or quality difference. If the application starts generation immediately on readiness, the example's 15-second interval disappears. A different reconnect path, service profile, duration limit or session end can change the arithmetic again.

Windflow's current rule is: Billing starts at the session's first generation_started event and stops when the session is released or ends. Queueing, worker startup, time before generation starts and the documented reconnect grace are not billed. After generation starts, paused or unwatched time remains billable; release the owner session to stop the meter. Reactor also documents continued ready-time charges while a recoverable disconnect retains the session. Disconnect or release deliberately and verify the terminal result; hiding the player is not a cost boundary.

## Scale usage without counting concurrency twice

For independent sessions, sum each session's billable seconds and multiply by its profile's rate. If 100 users each consume 150 Helios billable seconds at Windflow's checked rate, the usage calculation is `100 × 150 × $0.0017 = $25.50`. Whether those users overlap affects the capacity and concurrency you need; do not multiply the already-totaled 100 sessions by peak concurrency a second time.

Also distinguish **reserved balance** from **settled usage**. Windflow reserves the model rate multiplied by the maximum session ceiling at creation, then settles actual billable time and returns the unused reservation. Concurrent reservations can reduce spendable balance before that balance is fully consumed. The [pricing guide](https://windflow.online/guides/how-per-second-pricing-works) explains the lifecycle.

For a broadcast, establish whether several viewers share one owner session or receive separate independent sessions. Those are different capacity requirements. Retention, exported recordings, dedicated deployments and private delivery can introduce requirements outside a public model's metered rate; confirm the applicable arrangement before budgeting them.

## Get to cost per usable result

Run a small, representative evaluation before projecting thousands of sessions. Record the model/version, output settings, meter start and end, rejected starts, actual visible output and release confirmation. Include attempts that fail your acceptance criteria in the cost record rather than comparing only selected successful clips.

For live editing, inspect source/output quality and capture-to-visible delay. For a navigable world, use a repeatable route and control sequence. A lower per-second rate cannot compensate for missing required inputs or an output that does not meet the product's needs. Windflow's [SANA camera](https://windflow.online/guides/sana-camera-fps-and-latency), [JoyAI timing](https://windflow.online/guides/joyai-live-editing-latency-measurements) and [EdenLive compute](https://windflow.online/guides/edenlive-gpu-compute-measurements) reports illustrate different measurement boundaries; none measures a cross-provider price/performance winner.

Live Windflow use requires an approved private-beta account and credentials. A listed rate or model page does not guarantee that capacity is already running. Check current access, concurrency and the returned session-duration ceiling before scheduling traffic. For Reactor, Decart or fal, confirm the selected account's access, available tier and current model rate at the destination. Free credits and negotiated terms are not included in the table.

## FAQ

### Which realtime video API is cheapest?

There is no universal answer from this table. Match the model, task, quality requirements and billing window, then sum actual billable time across all attempts. The table's models and metering units are not interchangeable.

### Does per-second pricing mean per second of generated video?

Sometimes, but not always. Reactor documents ready-session time, Windflow meters after generation starts with its stated grace policy, and Decart distinguishes realtime active generation from video-job output seconds. Read the exact endpoint's unit.

### Does pausing a stream stop charges?

Not for a retained Windflow session after generation has started. Reactor likewise meters a session held in ready state, including documented recoverable disconnects. For other endpoints, confirm the specific pause and termination policy rather than inferring it from a per-second rate.

### Can I use an old Lucy article to estimate the current fal endpoint?

Use the model/version and price shown on the current endpoint. This check found Lucy 2.6 realtime at $0.02 per second on fal; a prior Lucy 2.5 article or a Decart-direct tier does not establish that endpoint's current account billing window.