Novita AI pricing
Novita AI’s RTX 4090 costs $0.33 an hour on demand and $0.17 as a spot instance (checked ). Pricing at Novita AI comes in two parts: GPUs rented by the hour, and model APIs paid per output. On its API, MiniMax H3 video costs $0.08 per second of video at 768P.
| Plan | Monthly price (USD) | Cost per second of 768P video (USD) | Paid ads |
|---|---|---|---|
| RTX 4090 | Pay per use | Not computed | Allowed with conditions |
| RTX 5090 | Pay per use | Not computed | Allowed with conditions |
| RTX 6000 Ada | Pay per use | Not computed | Allowed with conditions |
| H100 SXM | Pay per use | Not computed | Allowed with conditions |
| Storage | Pay per use | Not computed | Allowed with conditions |
| API: MiniMax H3 | Pay per use | $0.08 | Allowed with conditions |
Scroll sideways for all 4 columns
What you get on each plan, and notes
- RTX 4090
- What you get: 24 GB VRAM; on demand $0.33 an hour, spot $0.17 an hour, per GPU
- Notes: GPU page price table, read in the browser 12 November 2026. The spot page’s banner says
RTX 4090 from only $0.18/hour
the same day. - RTX 5090
- What you get: 32 GB VRAM; on demand $0.56 an hour per GPU; no spot price listed
- Notes: Same read.
- RTX 6000 Ada
- What you get: 48 GB VRAM; on demand $0.77 an hour, spot $0.39 an hour, per GPU
- Notes: Same read.
- H100 SXM
- What you get: 80 GB VRAM; on demand $3.39 an hour, spot $1.70 an hour, per GPU
- Notes: Same read.
- Storage
- What you get: Container disk 60 GB free, then $0.005 per GB a day; volume disk $0.005 per GB a day; network volume $0.002 per GB a day
- Notes: GPU instance pricing docs (last modified 5 November 2026).
- API: MiniMax H3
- What you get: Output video $0.08 a second at 768P, $0.13 a second at 2K; images beyond 5 inputs $0.04 each
- Notes: Pricing page, Video tab, read in the browser 12 November 2026. MiniMax H3 Max: output $0.05 a second at 480P, $0.08 at 768P. The pricing page’s Image filter lists no FLUX or Seedream model on 12 November 2026 (only Ming Image 0.1 Design, priced per million tokens).
Source: novita.ai/gpus
What one ad costs
Method: $0.08 per second of video at list price. Pricing page, Video tab (12 November 2026): MiniMax H3 output video at 768P costs 0.08 dollars a second, so a 6-second clip costs 0.48 dollars. 2K output costs 0.13 dollars a second. Input video is billed on top when you edit an existing clip.
A 6-second MiniMax H3 clip at 768P costs $0.48. The same clip at 2K costs $0.78. Ten hook variants at 768P come to $4.80. Whether H3 clips bought here can run in ads in the US, the EU, the UK or South Korea is unclear: see the licence note.
On a rented GPU there is no price per image to quote. You pay the hourly rate for the minutes the card runs. A two-hour session on the RTX 4090 costs $0.66 on demand or $0.34 on spot, plus storage.
Novita GPU prices per hour, on demand and spot
On-demand cards run until you stop them. Spot cards cost about half, but Novita can take them back (see the one-hour rule).
| GPU (memory) | On demand | Spot |
|---|---|---|
| RTX 4090 (24 GB) | $0.33 | $0.17 |
| RTX 5090 (32 GB) | $0.56 | Not published |
| RTX 6000 Ada (48 GB) | $0.77 | $0.39 |
| H100 SXM (80 GB) | $3.39 | $1.70 |
Scroll sideways for all 3 columns
Two Novita pages disagree on the RTX 4090 spot rate. The GPU page table lists the RTX 4090 spot rate at $0.17. On the spot page banner prices it “from only” $0.18 an hour. This page uses the table’s figure.
Memory decides which models fit on a card, so check a model’s size against the memory column before comparing rates.
To turn an hourly rate into a monthly budget for your ad volume, put it into the AI ad cost calculator.
What storage adds
Storage bills apart from the GPU, by the GB and by the day. Novita’s GPU pricing docs list three kinds:
- Container disk comes with 60 GB free, then $0.005 per GB a day.
- Volume disk costs $0.005 per GB a day.
- Network volume costs $0.002 per GB a day.
A 120 GB network volume comes to $0.24 a day. Over 30 days that is $7.20, whether a GPU runs or not.
When Novita’s billing starts and stops
The meter starts before you generate anything. Billing begins “when the instance is successfully created and enters the pulling state”, while it downloads its software image. It stops “when the instance is stopped”.
Billing Structure of On-Demand Pricing
Computing resources Cost = Instance Unit Price × Billing Duration × Number of Cards. Billing duration is accurate to the second and settled hourly. The billing start time is when the instance is successfully created and enters the pulling state, and the billing stop time is when the instance is stopped.
Container disk Pay-as-you-go. Supports 60GB free quota, charges for the excess based on capacity and usage duration. Unit price for the excess capacity: $0.005/GB/day.
Volume disk Unit price: $0.005/GB/day.
Network volume Unit price: $0.002/GB/day.
Time is “accurate to the second and settled hourly”. A large software image is billed GPU time while it downloads. So start from a template that already holds your tools, such as Novita’s official ComfyUI template. Stop the instance the moment a batch ends.
Novita AI API pricing for video
Novita’s pricing page prices video models per second of output.
| Model | 480P | 768P | 2K |
|---|---|---|---|
| MiniMax H3 | Not published | $0.08 | $0.13 |
| MiniMax H3 Max | $0.05 | $0.08 | Not published |
Scroll sideways for all 4 columns
Two charges sit on top. Input video is billed as well when you edit an existing clip. Input images beyond the first five cost $0.04 each.
Novita’s API reference calls this endpoint MiniMax H3 “overseas edition” and names no licence. MiniMax’s community licence, dated 2 August 2026, covers hosted APIs. Section V.4 bars using the model’s outputs “outside the Applicable Territory”, which excludes the US, the EU, the UK and South Korea. Novita does not say which document governs clips bought through it.
The same H3 clip on every other provider is priced in Open source video model API prices. Novita also sells language-model APIs priced per million tokens, on the same pricing page.
Spot instances: the one-hour rule
The spot page promises “Default 1-hour protection, 1-hour advance termination notice”. For the first hour the card is yours. After that, Novita can reclaim it with an hour’s warning.
That makes spot a fit for a batch of stills you can rerun. A render is protected for 1 hour, then reclaimable on 1 hour’s notice, so one that runs beyond about two hours is at risk. Save outputs to a network volume as you go.
Refunds and unused credits
Credits you load are spent or lost. Section 7 of Novita’s terms, updated 5 August 2026, says “All fees are non-refundable”, and section 8 adds that Novita “does not offer refunds for unused credits”.
Purchases and payment
All fees are non-refundable and exclusive of any applicable taxes, which the Customer is solely responsible for paying.
The exceptions are a verified outage that leaves the service “materially unavailable for an extended period”, a duplicate or incorrect charge, and refunds the law requires. Claims go to support within 30 days of the charge. Load a small balance first. What a new account gets free, and why no promo code exists, is in Novita AI promo code and free credits.
Novita beside Runpod and the other GPU clouds
Runpod’s RTX 4090 costs $0.34 an hour on Community Cloud and $0.89 on Secure Cloud, read . At list price on those read dates, Novita’s on-demand card sits a cent under Runpod’s Community Cloud rate.
Novita AI vs Runpod sets every card and rule side by side. Runpod pricing has Runpod’s full table, Runpod alternatives lists the other clouds and cloud GPU prices holds every rate we track. Novita AI covers its terms for ad work.
Questions buyers ask about Novita AI pricing
How much is an RTX 4090 on Novita?
$0.33 an hour on demand and $0.17 on spot, from the GPU page’s table, read .
Does Novita refund unused credits?
No. Section 8 of the terms makes all fees non-refundable, unused credits included. It allows refunds only for a verified outage, a billing error or where the law requires one.
What does Novita storage cost?
The first 60 GB of container disk is free. Extra container space and volume disks cost $0.005 per GB a day and a network volume $0.002.
Sources
- Novita AI: GPU instances, on-demand and spot prices per hour for the RTX 4090, RTX 5090, RTX 6000 Ada and H100 SXM. Checked
- Novita AI: Spot GPU instances, the banner pricing the RTX 4090
from only
0.18 dollars an hour, and the lineDefault 1-hour protection, 1-hour advance termination notice
. Checked - Novita AI: Pricing, Video tab, MiniMax H3 and H3 Max output rates per second by resolution and the charge for extra input images; the model APIs for language models sit on the same page. Checked
- Novita AI docs: GPU instance pricing, billing from the pulling state to the stop, counted to the second and settled hourly; container disk, volume disk and network volume rates. Checked
- Novita AI docs: MiniMax H3 API reference, names the endpoint MiniMax H3
overseas edition
and states no licence. Checked - Novita AI: Comfyui:flux1-fp8 template, official ComfyUI template with FLUX.1-dev, updated 31 March 2026. Checked
- Hugging Face, MiniMaxAI: MiniMax H3 Community License Agreement, licence date 2 August 2026; definition 5 (Excluded Territories) and section V.4. Checked
- Novita AI: Terms of Service, updated 5 August 2026; sections 7 (fees non-refundable) and 8 (refunds). Checked