How much do Novita AI dedicated GPUs cost?

Novita AI currently lists dedicated endpoint GPU rates of $0.61 per GPU-hour for RTX 4090, $1.99 for H100 SXM, and $2.99 for H200 SXM.

Does Novita AI charge for stopped dedicated endpoints?

Novita AI states that dedicated endpoints are billed only while replicas are running, with no charge when scaled to zero or stopped.

Who is Novita AI best for?

Novita AI is most relevant to developers and teams that need multiple AI model APIs, on-demand GPU infrastructure, or dedicated inference endpoints without managing the full infrastructure stack themselves.

Developer cloud buyer guide · Updated Sep 8, 2026

Novita AI Pricing: Serverless APIs vs Dedicated GPUs

Novita AI combines 200+ serverless model APIs, dedicated GPU endpoints, GPU instances and agent infrastructure in one platform. The important buying decision is not a single subscription tier — it is whether your workload fits usage-based APIs or dedicated compute.

Affiliate disclosure: COSHUMA may earn a commission from some links on this page, at no extra cost to you.

Try Novita AI → View official pricing

Current signup offer

Novita AI currently advertises $100 in Sandbox credits valid for 90 days →

See the COSHUMA free-credits buyer guide, what the offer does and does not imply, and how to use the credit period as a real cost test before scaling.

Serverless model APIs

Pay by model usage

Best when you want API access without reserving GPU capacity. Novita lists LLM, image, audio, video, vision and other models with model-specific usage rates.

Dedicated endpoints

From $0.61 / GPU-hour

Best for predictable or sustained inference. Running replicas are billed per second, and Novita states stopped or scale-to-zero replicas are not charged.

Spot GPU

Lower cost, interruptible

For interruption-tolerant workloads, Novita also advertises Spot GPU capacity with discounts versus on-demand pricing.

Dedicated endpoint GPU pricing

Current prices shown on Novita AI's official dedicated endpoint page:

GPU	VRAM	Price / GPU-hour	Good fit
NVIDIA RTX 4090	24 GB	$0.61	Smaller open models, cost-sensitive inference
NVIDIA H100 SXM	80 GB	$1.99	Larger models and higher-throughput workloads
NVIDIA H200 SXM	141 GB	$2.99	Very large models and memory-heavy inference

Prices can change. Verify the official Novita AI pricing page before deployment or purchase.

Choose serverless APIs when…

✓ Traffic is variable or early-stage.
✓ You want many models behind one API without managing GPUs.
✓ You prefer usage-based model pricing over reserved capacity.
✓ You are testing multiple LLM, image, audio or video models.

Choose dedicated endpoints when…

✓ You need consistent GPU-backed inference.
✓ Your workload runs long enough that dedicated capacity is easier to predict.
✓ You need OpenAI-compatible endpoint deployment for your own model.
✓ Scale-to-zero billing is useful during idle periods.

COSHUMA verdict

Novita AI is worth shortlisting when you want one vendor for both model APIs and GPU-backed deployment. For experimentation and uneven traffic, start with serverless APIs. For sustained inference or models you control, compare the dedicated GPU hourly cost against your expected utilization before committing.

Open Novita AI with COSHUMA referral → Check dedicated GPU rates

Frequently asked questions

How does Novita AI pricing work?

Serverless model API pricing depends on the specific model and usage. Dedicated endpoints use per-second billing on running GPU replicas.

What are the current dedicated GPU rates?

Novita currently lists RTX 4090 at $0.61/GPU-hour, H100 SXM at $1.99/GPU-hour, and H200 SXM at $2.99/GPU-hour.

Do idle dedicated endpoints cost money?

Novita says billing applies only to running replicas, with no charge when an endpoint is stopped or scaled to zero.

Does Novita AI have one flat monthly plan?

Not as the main buying model. The platform is primarily usage- and infrastructure-priced, so the right comparison is model/API usage versus GPU compute rather than a simple SaaS seat plan.

Represent this product?

Request a COSHUMA sponsored placement

Sponsored placement starts at USD 49. Approved sponsorships receive a clearly labeled promotional placement in designated high-visibility areas. Premium positions are priced separately based on placement and availability. Sponsorship does not change independent editorial ratings or organic rankings.

See sponsorship options → Email a sponsorship request →

This inquiry link does not create a charge.

Verification & sources

Last verified

September 8, 2026

Sources checked

novita.com

Software details and prices can change. Verify final terms on the vendor site before purchasing.

Practical value

Problems it can help solve

These are workflow problems the tool's documented capabilities can help address. COSHUMA does not promise a specific business result.

✓Developers need access to text, image, audio, video and vision models without separately operating model-serving infrastructure for each modality.
✓Teams moving from prototypes to production need a path from serverless model APIs to dedicated endpoints, GPU capacity and isolated workloads without rebuilding the application around a new provider.
✓AI-agent builders need hosted model access and sandbox-style infrastructure so they can test and run agent workloads without maintaining their own GPU fleet.

Service ideas

Ways to make money with this tool

Use the tool to create a service, workflow or deliverable someone may pay for. These are use-case ideas, not income guarantees.

Build multimodal AI prototypes for clients

Difficulty: Intermediate · Starting cost: Usage-based

Example deliverable: Working prototype that uses selected text, image, audio or video model APIs for one documented client workflow

Implement production model-API backends

Difficulty: Advanced · Starting cost: Usage-based

Example deliverable: API-backed application workflow with model selection, request handling, usage controls and deployment documentation

Offer AI-agent infrastructure setup