Sonic Inference Engine
The Sonic Inference Engine serves more than 400,000 image, video, audio, text and 3D models, hosted, partner, community or uploaded, through one task-based API on Runware's own hardware.
Find alternatives to Sonic Inference Engine
What it does
- Image generation
- Video generation
- Model inference
Features 5
- Intelligent per-request routingEach request is routed to a pod chosen by its load, last-token latency and cold-start state.
- Shared Model LakePods draw on one shared pool of models, with weights loaded into GPU memory on demand.
- Model upload APIYour own model weights are uploaded through the same API and become routable once the upload completes.
- Bring-your-own-container serverlessCustomers deploy their own container with custom code, models and runtimes on serverless compute.
- Multi-GPU large-model inferenceLarge models are served by distributing inference across several GPUs.
Sources
- Pricing
-
Pricing page: "Runware uses pay-as-you-go billing."
- Description
-
Official documentation · 10 Nov 2026
Product page: "The Sonic Inference Engine is a vertical inference platform: custom hardware and software engineered as one system."
- What it does
-
Official documentation · 10 Nov 2026
Product page: "The Sonic Inference Engine is a vertical inference platform: custom hardware and software engineered as one system."
- Sold within
-
Sold on its own with published plans.
- Name
-
Official documentation · 10 Nov 2026
Name as published on the product page.
- URL
-
Official documentation · 10 Nov 2026
Product page, read 2026-11-10.
- Deployment
-
Official documentation · 10 Nov 2026
Product page: "400K+ models, one addressing scheme: hosted, partner, community or your own upload."
- Interfaces
-
Official documentation · 10 Nov 2026
Product page: "400K+ models, one addressing scheme: hosted, partner, community or your own upload."
Also from Runware
-
Runware Serverless Infrastructure service
Serverless GPU service in which a customer brings code, a container or model weights and Runware handles GPU scheduling, workers and autoscaling from zero to production capacity, billed by the second.
-
Runware Compute Infrastructure service
Reserved GPU capacity on Runware's Sonic Pods as bare-metal nodes or full clusters, or as serverless GPU compute, in the US and Europe, with starting rates by GPU model and term.
-
Runware Model Router API service
Router that picks a model for each request from Runware's official catalog when the request sets the model to auto, ranked by the caller's weights for quality, speed and cost and an optional price ceiling.
-
Runware LoRA Training API service
Trains a style LoRA from a customer's uploaded images and registers it to the customer's account for generation through the same Runware API, priced per training step and with the trained weights downloadable.
-
Runware Model Serving API service
Hosts a customer's own model as a dedicated API endpoint, with Runware handling hosting, scaling and operations, billed per inference.
Something wrong here? Send a correction — quote this product id: runware-sonic-inference-engine.