All platforms compared on the basis of cost per minute of real-time interaction. Hidden costs (TTS, LLM, infrastructure) are noted separately.
Interactive tool
Cost Simulator
Interactive tool
Cost Simulator — Streaming Video Avatar
Estimate your monthly bill based on interaction volume. Main use case: real-time conversational video avatar.
Hosting mode
Usage Volume
Set users, session duration, parallel streams and the number of video avatars to train. Total minutes are computed automatically.
Open-source models (MuseTalk, LivePortrait, V-Express, etc.) have no per-minute price. Their total is a smoothed Exoscale hosting estimate: the model’s GPU, one instance per requested stream (1 stream/GPU, unmeasured), 160 h per month while the load fits that window, then the hours the load requires up to 720 h. Voice, LLM and operations stay separate.
Users / month250
1050010k
Avg session duration5 min
1 min10 min60 min
Parallel streams5 streams
1 stream20 streams500 streams
Video avatars4 avatars
1 avatar4 avatars6 avatars
Distinct characters to train. Où est Ava ? uses 4. From 1 to 6.
Users/month
250
Avg session
5 min
Total minutes/month
1.3k
250 × 5 min
Parallel streams
5
simultaneous capacity
Video avatars
4
to train
Formula: Volume = 250 users × 5 min = 1,250 min/month, 5 parallel streams, 4 video avatars. Each card then applies its vendor formula. An offer that does not cover duration, concurrency or the custom-face count is not selected. On Tavus, the first paid tier includes 1 face, then 3, 7 and 15.
Add external voice when it is not inside the minute ($0.046/min)
Filter by use case
17 platforms — sorted by total cost
BaselineSubscription, or the GPU window kept on infrastructure we run
UsageMinutes or hours beyond that baseline
Add-onVoice or stack added on top of the price
Same color on every offer. Length compares totals in dollars; the amount on the right stays in the offer’s currency.
Essential · Each session rounds up to the next minute. LITE minutes = credits; FULL minutes = credits ÷ 2. The cheapest open tier is kept.Business states 1 custom 1080p avatar. Extra slots are an add-on with no public price, so they are not added.· ≤ 20 min/session
Stream limits not documented — contact provider for high volumes.
Dev PAYG, no subscription · docs.dev.runwayml.com: gwm1_avatars = 2 credits then 2 credits every 6s, at $0.01 per credit. The Characters documentation caps a session at 5 min; the simulator excludes longer sessions instead of presenting them as continuous.· ≤ 5 min/session
Growth · Billed minutes = users × max(30s, rounded to 6s). Then the cheapest CVI tier that also covers custom faces: Free 0, Starter 1, Builder 3, Growth 7, Business 15. Free and Starter stop at their allowance.4 video avatars rule out Free (0), Starter (1), Builder (3). Selected tier: Growth (7 custom faces).
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × GPU3 RTX 3080 Ti (12 Go) × 160 h × CHF 1.0453. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × GPU3 RTX 3080 Ti (12 Go) × 160 h × CHF 1.0453. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Restrictive licence: this CHF does not authorize deployment. Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × A30 (24 Go) × 160 h × CHF 1.22645833. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × A30 (24 Go) × 160 h × CHF 1.22645833. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × A30 (24 Go) × 160 h × CHF 1.22645833. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × A40 (48 Go) × 160 h × CHF 1.34028. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × A40 (48 Go) × 160 h × CHF 1.34028. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
Stream limits not documented — contact provider for high volumes.
Smoothed Exoscale hosting · Smoothed estimate, not a measurement. 5 × RTX Pro 6000 (96 Go) × 160 h × CHF 2.15277778. Working window of 160 h while the load fits, otherwise session hours, capped at 720 h. 1 stream/GPU unmeasured. Exoscale small rate reread 19 October 2026.Restrictive licence: this CHF does not authorize deployment. Smoothed estimate. Software is 0. The CHF is GPU hosting for this load, not a per-minute price.
No public tier covers session length, concurrency, video avatars and volume together. The rest is a quote.
* Rate cards reread on 2026-10-19, except Runway reread on 2026-10-28. The cloud total follows the cheapest public tier that covers the session and the streams. Open source adds smoothed Exoscale hosting (160 h, then the load, capped at 720 h): that is not a vendor price. Simli past 50 min, Runway past 5 min and D-ID steps without a displayed price stay unpriced. External voice, if checked: $0.046/min (GPT-5 mini + Deepgram Flux TTS). 16 October 2026 rates used to sort euros and francs: €1 = $1.1253, CHF 1 = $1.2137.
Detailed comparison table
Click headers to sort
Filter platforms, then open the relevant profile.
11 platform(s) shown
Simli (Trinity-1)
Upd
Trinity-1 / Gaussian Splatting
GW 9/10
$/min
$0.000/min
Latency
~300ms
Sovereignty
3/5
Deployment
Cloud
19 octobre 2026 : 50 min/mois et 10 $ à l’inscription. La page /pricing répond 404, donc aucun $/min payant n’est affiché.
Docs du 17 octobre 2026 : $1 = 100 crédits. Cloud 4 crédits/min = $0.04 au top-up. Self-host parole 2 crédits = $0.02. Les dépassements $0.044 / $0.04 / $0.024 / $0.016 ne sont plus publiés. L’API Free est refusée dès le 19 octobre 2026 00:00 UTC.
LiveAvatar (Custom/LITE ou FULL) · HeyGen Voice 1 / Turbo · Avatar V · HeyGen Video · Video Agent · Cinematic
GW 8/10
$/min
$0.090/min
Latency
~300ms
Sovereignty
1/5
Deployment
Cloud
LiveAvatar relu le 18 octobre 2026 : Essential $99 / 1 100 crédits, dépassement $0,095 ; Business $475 / 6 000, $0,09. LITE = 1 crédit/min, FULL = 2. Le Starter n’est plus sur la page crédits. Le comparateur affiche la minute incluse Essential ($0,09) ; le dépassement LITE reste $0,095. HeyGen Voice et l’API vidéo async sont d’autres factures. Le <300 ms est un TTFF annoncé.
Grille officielle du 17 octobre 2026 en euros, pas les dollars précédents. Free 40 min. Starter €49 / 280 min / €0,175. Growth €149 / 1 490 min / €0,10. Scale €349 / 4 000 min / €0,0875. Enterprise jusqu’à €0,03/min. Le nombre trié reste l’ancien repère ; lire la note.
Docs Runway Dev relues le 28 octobre 2026 : GWM-1 Characters coûte 2 crédits à l’ouverture, puis 2 crédits par tranche de 6 s ; 1 crédit = 0,01 $. Une session complète de 5 min vaut donc 102 crédits = 1,02 $ (0,204 $/min). La grille Dev ne publie pas de palier d’abonnement qui réduise ce prix. Runway mentionne 37 ms de temps modèle et 1,75 s de turnaround serveur, pas une mesure TTFR ou de dialogue bout-à-bout comparable.
Grille du 19 octobre 2026 : Starter 8 $/41 min, dépassement 0,22 $. Creator 0,19 $, Ultra 0,18 $. Le slogan 0,048 $/min n’est pas une ligne de cette grille.
$0.367/min effectif au premier plan payant du bloc « Pricing Built to Scale » encore publié le 17 octobre 2026 (Starter $22 / 60 min). Builder $59 / 175 min, Growth $397 / 1 300 min. La même page expose un second bloc développeur (Starter $59 / 100 min, dépassement $0.37/min ; Growth $397 / 1 250 min, $0.32/min). Le portail garde le bloc complet, sans mélanger les deux grilles.
Docs du 17 octobre 2026 : $1 = 100 crédits. Cloud 4 crédits/min = $0.04 au top-up. Self-host parole 2 crédits = $0.02. Les dépassements $0.044 / $0.04 / $0.024 / $0.016 ne sont plus publiés. L’API Free est refusée dès le 19 octobre 2026 00:00 UTC.
LiveAvatar (Custom/LITE ou FULL) · HeyGen Voice 1 / Turbo · Avatar V · HeyGen Video · Video Agent · Cinematic
from $0.090/min
LiveAvatar relu le 18 octobre 2026 : Essential $99 / 1 100 crédits, dépassement $0,095 ; Business $475 / 6 000, $0,09. LITE = 1 crédit/min, FULL = 2. Le Starter n’est plus sur la page crédits. Le comparateur affiche la minute incluse Essential ($0,09) ; le dépassement LITE reste $0,095. HeyGen Voice et l’API vidéo async sont d’autres factures. Le <300 ms est un TTFF annoncé.
Grille officielle du 17 octobre 2026 en euros, pas les dollars précédents. Free 40 min. Starter €49 / 280 min / €0,175. Growth €149 / 1 490 min / €0,10. Scale €349 / 4 000 min / €0,0875. Enterprise jusqu’à €0,03/min. Le nombre trié reste l’ancien repère ; lire la note.
Docs Runway Dev relues le 28 octobre 2026 : GWM-1 Characters coûte 2 crédits à l’ouverture, puis 2 crédits par tranche de 6 s ; 1 crédit = 0,01 $. Une session complète de 5 min vaut donc 102 crédits = 1,02 $ (0,204 $/min). La grille Dev ne publie pas de palier d’abonnement qui réduise ce prix. Runway mentionne 37 ms de temps modèle et 1,75 s de turnaround serveur, pas une mesure TTFR ou de dialogue bout-à-bout comparable.
Grille du 19 octobre 2026 : Starter 8 $/41 min, dépassement 0,22 $. Creator 0,19 $, Ultra 0,18 $. Le slogan 0,048 $/min n’est pas une ligne de cette grille.
$0.367/min effectif au premier plan payant du bloc « Pricing Built to Scale » encore publié le 17 octobre 2026 (Starter $22 / 60 min). Builder $59 / 175 min, Growth $397 / 1 300 min. La même page expose un second bloc développeur (Starter $59 / 100 min, dépassement $0.37/min ; Growth $397 / 1 250 min, $0.32/min). Le portail garde le bloc complet, sans mélanger les deux grilles.
LemonSlice is the only platform offering two deployment modes at radically different costs: Hosted API ($0.21/min, zero infrastructure) and Self-Managed Pipeline (~$0.03/min on A100, full sovereignty). In Self-Managed mode, LemonSlice becomes the cheapest option after Simli and bitHuman, while being the only one to support multi-style avatars (cartoons, mascots, animals).
Most affordable
Simli ($0.009/min) and bitHuman Essence ($0.01/min) offer the best price-performance ratio for real-time video rendering.
Best sovereignty
bitHuman (on-premise CPU) and Beyond Presence (Munich, EU) are the only options with sovereign European deployment.
Best balance
Anam.ai (180ms, $0.12/min) and Beyond Presence (250ms, $0.175/min) offer the best latency/cost/sovereignty balance.
Note on real costs
The displayed $/minute cost covers only the avatar video rendering layer. For platforms without integrated LLM/TTS (Simli, bitHuman), add the same references used by the Voice Pipeline simulator: GPT-5 mini LLM $0.001/min and Deepgram Flux TTS $0.045/min. Quotes, discounts, and Enterprise costs must still be confirmed case by case.