HeyGen
How a leading AI video platform unlocked new product capabilities by replacing expensive managed providers with a single, cost-effective inference API.
Baseten alternative
The Baseten alternative: serverless GPUs and LLM APIs for less.
Deploy your own models on autoscaling GPUs billed per second, or call open LLMs per token, on infrastructure we design and run ourselves.
From $2.45H100 per GPU-hour, reserved
39% lowerH100 Serverless, no commitment
Up to 68%Lower on open LLMs
Baseten pricing vs Runware Serverless, per GPU-hour
Runware Serverless, with no commitment, is 39–40% lower than Baseten. Reserved capacity, on a 1 to 24 month term, is up to 62% lower.
Baseten dedicated-deployment prices from baseten.co/pricing (Basic plan), 11 Oct, 2026; Baseten publishes per-minute rates and the hourly figures are those × 60. Runware Serverless and reserved prices are our published Serverless pricing. Runware reserved rates require a 1–24 month term. Runware bills per second. GPU prices fluctuate, so list rates are subject to change.
Also on Runware: L40s, RTX PRO 6000, H200 and B300, from $0.63 per GPU-hour reserved.
Worked example
One H100 for a month, around the clock
What 730 hours on a single H100 costs per month, at the list rates in the table above. Reserve a baseline and burst above it on Serverless to land between the two Runware figures.
$4,745Baseten
$2,884Runware Serverless
$1,789Runware reserved
Baseten pricing vs Runware, per 1M tokens
Runware is lower on 5 of the 7 models both of us serve, by up to 68% on gpt-oss-120b, and level on GLM-5.3-Flash and Kimi K3.
Price view
Baseten Model API prices from baseten.co/pricing, 11 Oct, 2026. Runware prices from runware.ai model pages, 21 Oct, 2026. USD per 1M tokens.
Every open LLM we serve
The full LLM price list, and how the API works
The tables here cover the models both platforms list. Every language model on Runware, with its per-token rate, is on the pricing page.
Why teams switch
A Baseten alternative for serverless GPUs and LLM APIs
Lower price per GPU-hour and per token is the reason people look. What they stay for is a platform we run end to end, from the data centre to the autoscaler, with the open models on the same account.
Migration
How to move a Baseten deployment to Runware
A Baseten model is a container, and so is a Runware Serverless app. Bring the image, declare its endpoints in a container.yaml, and keep the GPU and scaling settings on the deploy command instead of in the package.
- Step 01
Bring your container
Truss packages a Baseten model as a Docker image. Bring that image, or any container that binds a port and answers POST per endpoint, with a container.yaml beside the Dockerfile that declares the port and the endpoints. The image runs your server exactly as it does on your machine.
- Step 02
Deploy with the GPU and worker limits you have today
GPU type, minimum and maximum workers and idle timeout are flags on the first deploy. Leave --min-workers at 0 to scale to zero, and change any of them later on a live app without a rebuild.
- Step 03
Point traffic at the new endpoint
Send a taskId with each request so retries are safe, watch workers, logs and cost in the console, and wind the Baseten deployment down when you're happy.
- Migrate to RunwareServerless quickstartBringing a container
container.yamlRunware Serverless
configVersion: 1
port: 8080
endpoints:
- path: generate
input:
type: object
required: [prompt]
properties:
prompt: { type: string }
# probes and timeouts take their defaults when omittedDeploy
runware serverless deploy --id my-model --gpu-type h100 \
--container . --min-workers 0 --max-workers 4
Powered by Sonic Pods
Why the price is lower: we build the data centre too.
Runware Serverless runs on Sonic Pods, modular AI data centres we design, build and operate ourselves. Up to 1,200 directly liquid-cooled GPUs in a 20 ft container, with no oversized building, no overbuilt redundancy and no years of sunk cost in front of them. Every component inside exists to serve inference, so you pay for GPU time rather than someone else's overhead.
- Cheaper buildout vs traditional
- 100×
- Power efficiency, liquid cooled
- 95%
- To deploy a pod, not years
- Days
No trainingOn your data
SSO + SAMLEnterprise auth
SOC 2Certified
ISO 27001Certified
GDPRCompliant
24/7Engineering support
Teams that switched and scaled
Lower bill, no GPU platform to run. Here's what moving to Runware looked like for teams shipping at production scale.
Are running with us
FAQ
Baseten alternative, answered
Yes, for both halves. Runware Serverless runs your own containers or Python on GPUs that scale from zero to production capacity and back, billed per second, which is the shape of a Baseten dedicated deployment. The same account calls open LLMs per token through an OpenAI-compatible endpoint, which is the shape of Baseten's Model APIs, and adds 400K+ image, video, audio and 3D models.
An H100 is $6.50 per hour on Baseten and $3.95 on Runware Serverless, 39% lower with no commitment on either side; across the 2 GPUs both of us sell the saving is 39–40%. Reserved, on a 1 to 24 month term, the same H100 is as low as $2.45, 62% lower. Baseten prices are from baseten.co/pricing on 11 Oct, 2026, converted from per-minute to per-hour; Runware prices are our published Serverless rates.
Bring the container. Truss packages a Baseten model as a Docker image, and Runware Serverless deploys any image that binds a port, answers POST for each endpoint and serves a readiness route, declared in a container.yaml beside the Dockerfile. GPU type and worker limits are flags on the deploy. The container guide covers the contract, and our engineers help teams migrate.
Of the models on Baseten's Model API list, Runware serves gpt-oss-120b, DeepSeek-V4.1-Flash, GLM-5.2, DeepSeek-V4-Flash-0731, GLM-5.3, GLM-5.3-Flash and Kimi K3 per token through an OpenAI-compatible endpoint, with Claude, Gemini and GPT on the same key. GLM-5.3 Fast, GLM-5.2 Fast, DeepSeek V4.1 Flash Fast, DeepSeek V4 Pro 0813 and NVIDIA Nemotron 3 Ultra are not on Runware today; if you need one of them, get in touch. The full list is in the model directory.
No. Serverless bills by the second while a worker is allocated to your app. With the minimum worker count at zero an idle app costs nothing, and time spent waiting for a GPU, pulling your image or recovering from a failure on our side is not metered. Holding workers warm with a minimum worker count is billed at the Serverless rate for as long as they are held.
Serverless draws workers from the shared pool and bills by the second while a worker holds a GPU, with no commitment. Reserved commits you to a number of GPUs of a given type for a 1 to 24 month term at a lower rate, and those GPUs stay yours even when the pool is full. Anything above the reservation bursts into Serverless. Most production teams reserve a baseline and burst above it.
Reserved capacity
Running at scale? Talk to us about reserved capacity.
Reserved GPUs start as low as $2.45 per GPU-hour on a 1 to 24 month term, and anything above the reservation bursts into Serverless. Volume token pricing is a conversation too.