Pay only for what's running
Per-second billing on active replicas. Scale to zero, pay zero. No charges for idle endpoints, no minimum commitments.
Billing preview
3 replicas 4090 × 2h 15m 42s running
= 3 × 8,142s × $0.000608/s
= $14.85
DEDICATED ENDPOINT
Run Your Models,
We Handle the Rest
Deploy models on dedicated GPUs. Per-second billing on running replicas only, affordable GPU options starting at $0.61/hr, and inference issues are on us.
GPUs for every budget
Not every workload needs an H200. Choose from RTX 4090, RTX 5090, H100 — match the GPU to your model size and your budget.
RTX 4090
24 GB
from$0.61/hr
H100
80 GB
from$1.99/hr
H200
141 GB
from$2.99/hr
Inference breaks? That's on us.
OOM, CUDA errors, model loading failures — our team diagnoses and resolves. You get a clear explanation, not a cryptic stack trace.
Report via console or email
Our team investigates root cause
You get diagnosis + resolution
Deployment options
Serverless vs Dedicated
Choose the right deployment model for your workload
Serverless Endpoints
Pay per token- 600+ models available
- Zero infrastructure management
- Auto-scaling included
- Best for variable workloads
Recommended for production
Dedicated Endpoints
- Isolated GPU resources
- Guaranteed latency SLA
- Custom models & LoRA adapters
- Scale-to-zero support
- Best for predictable, high-throughput workloads
Platform
Built for Production Workloads
Everything you need to deploy, scale, and manage AI inference in production.
Blazing-Fast Inference
Powered by optimized serving engines on NVIDIA H200, H100, and RTX 4090 GPUs. Sub-second latency for real-time applications.
Dynamic Auto-Scaling
Scale from 0 to N replicas automatically based on traffic. Set min/max replicas and scale-down delay to match your traffic patterns.
LoRA Adapter Support
Hot-swap LoRA adapters on running endpoints without restarts. Deploy multiple fine-tuned variants on a single base model.
Flexible GPU Options
Choose from NVIDIA H200, H100, and RTX 4090. Select tensor parallelism and GPU count to match your model requirements.
Pay-Per-Hour Billing
Billed by GPU-hour with per-second granularity. Scale to zero when idle — no minimum commitment, no idle GPU costs.
Enterprise Support
Dedicated technical support, custom SLAs, and priority access to new GPU types. Volume discounts for large deployments.
Catalog
Deploy Popular Open-Source Models
One-click deploy the most popular LLMs, or bring your own Hugging Face model.
Workflow
Deploy in 3 Steps
Everything you need to deploy, scale, and manage AI inference in production.
001
Pick your model
Search 50K+ models from Hugging Face, or paste your private repo URL.
Search models... e.g. Qwen2.5-7B
Qwen2.5-7B-Instruct
Llama-3.1-8B
Mistral-7B-v0.3
002
Choose your GPU
See the recommended GPU for your model. Pick the one that fits your budget.
RTX 4090 · $0.61/hr
H100 · $1.99/hr
H200 · $2.99/hr
Replicas
1
003
Deploy
Your endpoint is live in minutes. OpenAI-compatible URL ready to use.
Endpoint
Live
POST
api.example.com/v1/chat/completions
Pricing
Transparent GPU Pricing
| GPU | VRAM | Price / GPU-hour |
|---|---|---|
NVIDIA H200 SXMPopular | 141 GB | $2.99 |
NVIDIA H100 SXM | 80 GB | $1.99 |
NVIDIA RTX 4090 | 24 GB | $0.61 |
Need reserved capacity or custom pricing?
Talk to our teamFAQ
Frequently Asked Questions
Per-second on running replicas only. When your endpoint is scaled to zero or stopped, you pay nothing. No minimum commitments, no idle charges.
Still have questions? Contact support
Everything you need to build production AI.
200+ models, on-demand GPUs, and secure agent runtimes — unified under one API. Free to start, scales as you grow.