1-Click Model Deployment

Same models. Same GPUs. Up to 85% less per hour.

Pick a model from our curated catalog and deploy it in one click on spot GPUs arbitraged across AWS, GCP and RunPod — the guaranteed <30s preemption recovery SLA keeps endpoints serving even on spot capacity.

Request Access See the model math

1-click

Deploys from a curated model catalog

<30s

Recovery SLA — benchmarked in CI

Clouds arbitraged — AWS · GCP · RunPod

730h

Per-month always-on pricing basis

GPU Rates

The same silicon, a fraction of the hourly rate.

Per-GPU hourly list rates, July 2026. AetherCompute rates are live spot-arbitrage estimates refreshed hourly.

		On AetherCompute	What you'd pay elsewhere
GPU	VRAM	AetherCompute	HF Inference Endpoints	Replicate	You save vs HF
RTX 4090	24 GB	$0.22	—	—	—
L40S	48 GB	$0.35	$1.80	$3.51	−81%
A100 SXM	80 GB	$0.65	$2.50	$5.04	−74%
H100 SXM	80 GB	$0.95	$4.50	$5.49	−79%
B200	192 GB	$1.85	$9.25	—	−80%

RunPod and Modal H100 list rates (~$2.69 and $3.95/hr respectively) sit between HF and AetherCompute. Per-second serverless platforms only win at low utilization — see methodology.

Model Hosting Costs

What 1-click deployment saves on the models you already use.

The curated launch catalog on always-on (730 hrs/mo) serving configs. GPU count × type shown per deployment.

		On AetherCompute		What you'd pay elsewhere
Model	Type	AetherCompute config	AetherCompute $/mo	HF Endpoints $/mo	Replicate $/mo	Savings
Llama 3.1 8B Instructmeta-llama/Llama-3.1-8B-Instruct	LLM	1× RTX 4090 · FP16	$161	$584	$2,562	−72%
Mistral Small 3 24Bmistralai/Mistral-Small-3-24B-Instruct	LLM	1× L40S · INT8	$256	$1,314	$2,562	−81%
Gemma 3 27B ITgoogle/gemma-3-27b-it	LLM	1× L40S · INT8	$256	$1,314	$2,562	−81%
DeepSeek-R1 Distill 32Bdeepseek-ai/DeepSeek-R1-Distill-Qwen-32B	LLM	1× A100 · FP16	$475	$1,825	$3,679	−74%
Mixtral 8×7B Instructmistralai/Mixtral-8x7B-Instruct-v0.1	LLM	2× L40S · FP16	$511	$2,628	$5,124*	−81%
Llama 3.3 70B Instructmeta-llama/Llama-3.3-70B-Instruct	LLM	2× A100 · FP8	$949	$3,650	$7,358	−74%
Qwen2.5 72B InstructQwen/Qwen2.5-72B-Instruct	LLM	2× A100 · FP8	$949	$3,650	$7,358	−74%
Llama 3.1 405B Instructmeta-llama/Llama-3.1-405B-Instruct	LLM	4× H100 · FP8	$2,774	$13,140	contract-only	−79%
FLUX.1-devblack-forest-labs/FLUX.1-dev	Vision	1× RTX 4090 · FP8	$161	$1,314	$2,562	−88%
SDXL 1.0stabilityai/stable-diffusion-xl-base-1.0	Vision	1× RTX 4090	$161	$584	$2,562	−72%
Whisper Large v3openai/whisper-large-v3	Audio	1× RTX 4090	$161	$584	$2,562	−72%

*Mixtral Replicate: 2× L40S × $3.51 × 730 = $5,124. Savings % are vs HF Inference Endpoints. Monthly = hourly × 730, 24/7 always-on.

How It Works

From model pick to live endpoint in three steps.

No Dockerfiles, no GPU hunting, no spot anxiety.

Pick a model from the catalog

A curated set of production-ready open models. Weights hydrate instantly over the FUSE layer — no multi-GB stage-in wait.

We pick the cheapest GPU config

Hourly arbitrage across AWS Spot, GCP preemptible and RunPod places the model on the cheapest compliant $/GPU-hr, with GPU bin-packing for multi-tenant efficiency.

Spot-proof serving

If a node is preempted, checkpoint + failover migration restores the endpoint in under 30 seconds — guaranteed by a CI-benchmarked SLA.

Methodology & assumptions

Competitor figures are public list prices as of July 2026 (sources: huggingface.co/pricing, replicate.com/pricing); HF Inference Endpoints AWS region rates; multi-GPU HF configs scale linearly from published 1×/4×/8× rates.
Replicate per-second rates converted to hourly equivalents assuming 100% utilization — bursty low-traffic workloads can cost less on per-second billing.
AetherCompute rates are illustrative spot-arbitrage rates that vary by region and time.
Monthly basis: 730 hours always-on, single replica, vLLM serving with the quantization shown. Models fit the stated GPU config in the stated precision; configs are illustrative minimums.