Serverless Python GPU — more GPUs, simpler pricing
Velar and Beam are in the same category: serverless Python GPU runtimes with scale-to-zero. Velar offers a broader GPU selection — from L4 at $0.49/hr to H200 — with free starter credits and straightforward per-second billing.
Why developers choose Velar over Beam
If you are already on Beam or evaluating it, Velar offers the same serverless Python workflow with a wider GPU catalogue and competitive pricing from L4 to H200.
Broader GPU selection, competitive rates
Velar offers 7 GPU tiers from L4 at $0.49/hr to H200 at $4.49/hr — pick the right hardware for your workload and budget, not just what the platform stocks.
Same serverless Python workflow
If you know how to use Beam, you already know how to use Velar. Decorate a function, specify a GPU, deploy. The developer experience is intentionally familiar.
Persistent endpoints included
Spin up always-on GPU endpoints for latency-sensitive applications. Combine with serverless functions for the right tool per workload — all in one platform.
Full GPU catalogue on Velar
Per-second billing on every GPU. Scale to zero when idle.
| GPU | Velar Price | Best for |
|---|---|---|
| L4 24GB | $0.49/hr | Small LLMs, image gen |
| RTX 4090 24GB | $0.74/hr | Fast inference, fine-tuning |
| L40S 48GB | $1.08/hr | SDXL, medium LLMs |
| A100 80GB | $1.74/hr | Large LLMs, training |
| H100 PCIe | $2.99/hr | Production inference |
| H100 SXM | $3.36/hr | Max throughput |
| H200 | $4.49/hr | Largest models |
Pricing as of May 2026. Check velar.run/pricing for current rates.
Velar vs Beam
| Feature | Velar | Beam |
|---|---|---|
| Python-native SDK | ||
| Scale to zero | ||
| Per-second billing | ||
| No Dockerfile required | ||
| Persistent endpoints | ||
| L4 GPU availableVelar offers L4 at $0.49/hr — ideal for cost-efficient inference | ||
| H100 SXM availableVelar offers H100 SXM at $3.36/hr for maximum throughput | ||
| H200 availableVelar offers H200 at $4.49/hr | ||
| Free starter credits |
Familiar serverless Python API
The decorator-based workflow is identical in concept to Beam. If you are migrating, the refactor is straightforward — change the import, update the decorator, redeploy. Your Python logic stays the same.
- pip install velar-sdk and you are ready
- Same scale-to-zero serverless model
- Works with any Python ML library
- Persistent endpoints for always-warm inference
import velar
app = velar.App("inference-api")
image = velar.Image.from_registry(
"pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime"
).pip_install("transformers", "accelerate")
@app.function(gpu="L4", image=image)
def classify(text: str) -> dict:
from transformers import pipeline
classifier = pipeline(
"text-classification",
model="cardiffnlp/twitter-roberta-base-sentiment-latest",
device=0,
)
return classifier(text)[0]
# Fan out over a dataset in parallel
app.deploy()
texts = load_dataset() # your data
results = [classify.remote(t) for t in texts]Frequently asked questions
How does Velar compare to Beam for serverless Python workloads?
Both Velar and Beam offer serverless GPU Python runtimes with scale-to-zero and per-second billing. The primary differences are GPU selection — Velar offers 7 GPU tiers including L4 at $0.49/hr and H200 — and pricing. Velar also offers free starter credits to evaluate before committing.
Does Velar support the same @app.function decorator pattern?
Yes. The workflow is identical in concept: decorate a Python function with your GPU and image config, then deploy. If you are migrating from Beam, the refactor is typically under an hour.
What is the cheapest GPU available on Velar?
The NVIDIA L4 24GB is available from $0.49/hr. It is well-suited for small-to-medium LLM inference, image generation with Stable Diffusion 1.5/2.1, and embedding generation workloads.
Can I run persistent endpoints, not just on-demand functions?
Yes. Velar supports both serverless functions (scale-to-zero) and persistent endpoints (always-warm, dedicated GPU). Use functions for bursty workloads and endpoints for latency-sensitive production APIs.
Is there a free tier?
Velar Cloud is paused and new accounts are not being created right now. Join the waitlist to be notified at relaunch.
Ready to make the switch?
Velar Cloud is paused. Join the waitlist for the relaunch. Deploy your first workload in under 5 minutes.