Live · refreshed hourly
Cloud GPU prices right now
What it actually costs to rent each GPU this hour, and further down, the cheapest card that genuinely runs each open model. Real offers from the providers we track, not a marketing rate card.
- GPUs priced
- 31
- live offers
- 231
- providers tracked
- 4
Captured
Before you rent
Rent a GPU, or just call an API?
If you only want answers from an open model, a per-token API can cost less than an hour of GPU time. If you need private data, fine-tuning or image and video models, rent the card.
Option A
Rent a GPU by the hour
For private data, fine-tuning, image and video models, or long sessions. You pay while it runs.
- 16–48 GBNVIDIA L40S · Vast.ai$0.04/hr
- 80 GB+NVIDIA A100 80GB PCIe · Vast.ai$0.15/hr
Cheapest listed hourly price per VRAM tier, updated 25m ago. 4 providers tracked.
Option B
Just call an API, pay per token
For chat, coding and agents on open models, with no GPU to manage. You pay for what you send and get back.
- OpenAI GPT OSS 120B$0.05 in · $0.25 out
- Llama 3.3 70B Instruct$0.135 in · $0.40 out
- Qwen3.6 35B A3B$0.248 in · $1.485 out
USD per 1M tokens, Novita AI's published rates, read Oct 29, 2026. Bars show the output price.
Cost to run any model, both waysLive price table
Every GPU we price, cheapest first
Minimum live $/hr per provider (median in grey when it differs). Spot is interruptible; on-demand is reserved while you run it.
| GPU | VRAM | Vast.ai · spot | Runpod · on-demand | Runpod · spot | Lium | Novita AI | Cheapest | Rent |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 3060 12GB | 12 GB | $0.03 · med $0.06 | — | — | — | — | $0.03/hr | Vast.ai |
| NVIDIA L40S | 48 GB | $0.04 | $0.59 | — | $0.33 · on-demand | $0.55 · on-demand | $0.04/hr | Vast.ai |
| NVIDIA Tesla V100 32GB SXM2 | 32 GB | $0.05 · med $0.08 | $0.19 | — | — | — | $0.05/hr | Vast.ai |
| NVIDIA RTX A4000 | 16 GB | $0.06 · med $0.09 | $0.25 | — | — | — | $0.06/hr | Vast.ai |
| NVIDIA GeForce RTX 4060 Ti 16GB | 16 GB | $0.06 · med $0.07 | — | — | — | — | $0.06/hr | Vast.ai |
| NVIDIA GeForce RTX 5060 Ti 16GB | 16 GB | $0.07 · med $0.13 | — | — | $0.22 · on-demand | — | $0.07/hr | Vast.ai |
| NVIDIA GeForce RTX 4070 | 12 GB | $0.07 · med $0.13 | — | — | — | — | $0.07/hr | Vast.ai |
| NVIDIA GeForce RTX 3090 Ti | 24 GB | $0.09 · med $0.16 | $0.46 | — | $0.16 · spot | — | $0.09/hr | Vast.ai |
| NVIDIA GeForce RTX 4070 SUPER | 12 GB | $0.09 | — | — | — | — | $0.09/hr | Vast.ai |
| NVIDIA GeForce RTX 3080 Ti | 12 GB | $0.09 · med $0.11 | $0.18 | — | — | — | $0.09/hr | Vast.ai |
| NVIDIA GeForce RTX 3080 12GB | 12 GB | $0.09 · med $0.13 | $0.17 | — | — | — | $0.09/hr | Vast.ai |
| NVIDIA GeForce RTX 4070 Ti SUPER | 16 GB | $0.10 · med $0.12 | $0.50 | — | $0.14 · on-demand | — | $0.10/hr | Vast.ai |
| NVIDIA RTX A5000 | 24 GB | $0.10 · med $0.22 | $0.27 | — | — | — | $0.10/hr | Vast.ai |
| NVIDIA GeForce RTX 5070 Ti | 16 GB | $0.11 · med $0.14 | — | — | $0.25 · on-demand | — | $0.11/hr | Vast.ai |
| NVIDIA GeForce RTX 4080 SUPER | 16 GB | $0.13 · med $0.19 | $0.50 | — | — | — | $0.13/hr | Vast.ai |
| NVIDIA A100 80GB PCIe | 80 GB | $0.15 · med $0.24 | — | — | — | — | $0.15/hr | Vast.ai |
| NVIDIA GeForce RTX 5080 | 16 GB | $0.18 · med $0.20 | $0.59 | — | $0.36 · on-demand | — | $0.18/hr | Vast.ai |
| NVIDIA Tesla T4 | 16 GB | — | — | — | $0.18 · spot | — | $0.18/hr | Lium |
| NVIDIA GeForce RTX 4090 | 24 GB | $0.20 · med $0.40 | $0.89 | — | $0.35 · on-demand | — | $0.20/hr | Vast.ai |
| NVIDIA GeForce RTX 5090 | 32 GB | $0.27 · med $0.56 | $1.19 | — | $0.71 · spot | — | $0.27/hr | Vast.ai |
| NVIDIA RTX 5000 Ada Generation | 32 GB | $0.31 | $0.83 | — | — | — | $0.31/hr | Vast.ai |
| NVIDIA RTX 6000 Ada Generation | 48 GB | $0.33 · med $0.47 | $0.99 | — | $0.80 · on-demand | — | $0.33/hr | Vast.ai |
| NVIDIA RTX A6000 | 48 GB | $0.33 · med $0.45 | $0.59 | — | $0.42 · spot | — | $0.33/hr | Vast.ai |
| NVIDIA H100 SXM | 80 GB | $0.78 · med $1.73 | $3.99 | — | $2.75 · spot | — | $0.78/hr | Vast.ai |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 96 GB | — | — | — | $1.29 · spot | — | $1.29/hr | Lium |
| NVIDIA H200 SXM | 141 GB | $1.58 · med $1.93 | $5.29 | — | $3.74 · spot | — | $1.58/hr | Vast.ai |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | 96 GB | — | $2.49 | — | $1.59 · on-demand | — | $1.59/hr | Lium |
| NVIDIA RTX PRO 6000 Blackwell Server Edition | 96 GB | — | — | — | $1.60 · spot | — | $1.60/hr | Lium |
| NVIDIA H100 PCIe | 80 GB | $1.67 | $2.89 | — | $1.75 · on-demand | — | $1.67/hr | Vast.ai |
| AMD Instinct MI300X | 192 GB | — | $2.39 | — | — | — | $2.39/hr | Runpod |
| NVIDIA B200 | 192 GB | $5.94 · med $7.34 | $7.99 | — | $5.94 · spot | — | $5.94/hr | Vast.ai |
Lium: decentralized GPU marketplace; spot nodes can be interrupted. Its column shows the cheaper of spot and on-demand nodes rentable as a single GPU right now; we list it as price data only — plain link, no referral. Novita AI's column comes from its public GPU list (about five single-GPU products); a card shows only while Novita reports it in stock, cheaper of spot and on-demand.
By model
Cheapest way to run each model in the cloud
The cheapest live-priced GPU each model fits on at 16k context, computed from the model's own config, and what a month of it costs.
| Model | Cheapest GPU that fits | Quant | $/hr | 8h/day · mo | 24/7 · mo |
|---|---|---|---|---|---|
| Kimi K3 (MoE) | Doesn't fit any single GPU we price at this context — see the model page for multi-GPU and unified-memory options. | ||||
| DeepSeek V4 Pro (MoE) | Doesn't fit any single GPU we price at this context — see the model page for multi-GPU and unified-memory options. | ||||
| DeepSeek V4.1 Flash (MoE) | Doesn't fit any single GPU we price at this context — see the model page for multi-GPU and unified-memory options. | ||||
| GLM-5.3 (MoE) | Doesn't fit any single GPU we price at this context — see the model page for multi-GPU and unified-memory options. | ||||
| MiniMax-M3 (MoE) | AMD Instinct MI300X · 192 GBRunpod | Q2_K169 GB | $2.39 | $574 | $1721 |
| Qwen3 235B-A22B (MoE) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition · 96 GBLium | Q2_K95 GB | $1.29 | $310 | $929 |
| MiniMax-M2.5 (MoE) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition · 96 GBLium | Q2_K94 GB | $1.29 | $310 | $929 |
| Qwen3.8-Flash-Next (MoE) | NVIDIA A100 80GB PCIe · 80 GBVast.ai | Q2_K71 GB | $0.15 | $35 | $106 |
| Qwen3.5 122B-A10B (MoE) | NVIDIA A100 80GB PCIe · 80 GBVast.ai | Q4_K_M72 GB | $0.15 | $35 | $106 |
| NVIDIA Nemotron 3 Super 120B-A12B (MoE) | NVIDIA A100 80GB PCIe · 80 GBVast.ai | Q4_K_M71 GB | $0.15 | $35 | $106 |
| gpt-oss 120B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q2_K47 GB | $0.04 | $10 | $29 |
| Llama 4 Scout 17B-16E (MoE) | NVIDIA L40S · 48 GBVast.ai | Q2_K44 GB | $0.04 | $10 | $29 |
| GLM-4.5-Air (MoE) | NVIDIA L40S · 48 GBVast.ai | Q2_K45 GB | $0.04 | $10 | $29 |
| Qwen3-Next 80B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q4_K_M46 GB | $0.04 | $10 | $29 |
| Qwen3-Coder-Next 80B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q4_K_M46 GB | $0.04 | $10 | $29 |
| DeepSeek R1 Distill 70B | NVIDIA L40S · 48 GBVast.ai | Q4_K_M46 GB | $0.04 | $10 | $29 |
| Mixtral 8x7B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q6_K38 GB | $0.04 | $10 | $29 |
| Qwen3.6 35B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| Ornith-1.5 35B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| Qwen3.5 35B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| Qwen2.5 Coder 32B | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| Qwen2.5 32B Instruct | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| DeepSeek R1 Distill 32B | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| Qwen3 32B | NVIDIA L40S · 48 GBVast.ai | Q8_037 GB | $0.04 | $10 | $29 |
| NVIDIA Nemotron 3 Nano 30B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_032 GB | $0.04 | $10 | $29 |
| Gemma 4 31B | NVIDIA L40S · 48 GBVast.ai | Q8_033 GB | $0.04 | $10 | $29 |
| Qwen3 30B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_032 GB | $0.04 | $10 | $29 |
| Qwen3-Coder 30B-A3B (MoE) | NVIDIA L40S · 48 GBVast.ai | Q8_032 GB | $0.04 | $10 | $29 |
| Qwen3.6 27B | NVIDIA L40S · 48 GBVast.ai | Q8_029 GB | $0.04 | $10 | $29 |
| Qwen3.8 27B | NVIDIA L40S · 48 GBVast.ai | Q8_029 GB | $0.04 | $10 | $29 |
| Qwen3.5 27B | NVIDIA L40S · 48 GBVast.ai | Q8_029 GB | $0.04 | $10 | $29 |
| Gemma 4 26B-A4B (MoE) | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q2_K11 GB | $0.03 | $8 | $24 |
| Mistral Small 24B | NVIDIA L40S · 48 GBVast.ai | F1647 GB | $0.04 | $10 | $29 |
| gpt-oss 20B (MoE) | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | IQ4_XS12 GB | $0.03 | $8 | $24 |
| DeepSeek R1 Distill 14B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q4_012 GB | $0.03 | $8 | $24 |
| Qwen2.5-Coder 14B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q4_012 GB | $0.03 | $8 | $24 |
| Qwen3 14B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q4_K_M12 GB | $0.03 | $8 | $24 |
| Phi-4 14B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q4_012 GB | $0.03 | $8 | $24 |
| Gemma 3 12B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q6_K11 GB | $0.03 | $8 | $24 |
| Gemma 4 12B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q6_K10 GB | $0.03 | $8 | $24 |
| Qwen3.5 9B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q8_011 GB | $0.03 | $8 | $24 |
| Qwen3 8B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q8_011 GB | $0.03 | $8 | $24 |
| Gemma 4 E4B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q8_09 GB | $0.03 | $8 | $24 |
| Qwen2.5 7B Instruct | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q8_09 GB | $0.03 | $8 | $24 |
| Mistral 7B Instruct v0.3 | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | Q8_010 GB | $0.03 | $8 | $24 |
| Gemma 4 E2B | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | F1610 GB | $0.03 | $8 | $24 |
| Qwen3 4B Instruct 2507 | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | F1611 GB | $0.03 | $8 | $24 |
| Llama 3.2 3B Instruct | NVIDIA GeForce RTX 3060 12GB · 12 GBVast.ai | F169 GB | $0.03 | $8 | $24 |
Monthly figures are the live $/hr × 240h (8h/day) or × 720h (24/7). Whether renting beats buying depends on your own usage — the rent-vs-buy break-even walks through it honestly.
Price alerts
Get a note when H100 or 5090 rentals drop
Prices have fallen 60–75% from their 2024 peak and still move week to week. We watch them hourly; you get the drops that matter, not the noise.
The Aliteq brief
The tech worth knowing — hardware, AI, gaming, deals. No spam, unsubscribe anytime.
Referral linkThe cheapest spot marketplace
A marketplace of independent hosts — usually the lowest $/hr anywhere, with reliability that varies by host. Best for interruptible work.
Referral linkManaged pods + serverless
Predictable on-demand pods, one-click templates and per-second serverless — the easiest way to go from zero to a running GPU.
New to Runpod? Sign up with Google through this link and load $10: Runpod adds a bonus credit, random between $5 and $500 outside Europe ($5 in Europe), valid 90 days.
Per-token model APIs + GPU instances
Call open models like Llama, Qwen, Gemma and gpt-oss by API and pay per token — no GPU to set up. It also rents GPU instances.
Lium
Lium: decentralized GPU marketplace; spot nodes can be interrupted. Price data only — no referral.
Other providers · not in the price table yet
Hourly GPU VMs + bare metal
GPU virtual machines rented by the hour (their site: "No contracts, no surprises"), plus single-tenant bare-metal servers.
FAQ
How to read this page
How often are these cloud GPU prices updated?
Hourly. Our worker pulls live offers from each provider's API and stores the minimum, 25th-percentile and median $/hr per GPU. The capture time is shown on the page — never quote a price without its timestamp.
Which cloud GPU providers are tracked?
4 with live prices right now, out of four we pull from: Vast.ai (a spot marketplace of independent hosts — cheapest, variable reliability), Runpod (managed on-demand pods and spot), Lium (a decentralized GPU marketplace; its spot nodes can be interrupted) and Novita AI (its public GPU list: about five single-GPU products, shown only while in stock). Vast.ai, Runpod and Novita AI links are referral links; Lium is listed as price data only, with a plain link and no referral. We add providers to the price table only where a live, legally usable price feed exists; we don't scrape. Massed Compute (hourly GPU VMs) is linked further down through a referral link, without prices in the table yet.
What's the difference between spot and on-demand GPU pricing?
Spot (interruptible) capacity is cheaper but can be reclaimed by the provider mid-job — fine for inference and checkpointed training. On-demand is reserved while you run it and costs more. The table shows both where a provider offers them.
How is 'cheapest GPU that runs a model' calculated?
From each model's own config (layers, hidden size, heads) we compute the VRAM it needs at each quantisation at 16k context (bits-per-weight constants validated to a 0.7% median error), then pick the cheapest live-priced GPU it fits on. It is an estimate of fit and cost, not a benchmark.
Disclosure:Vast.ai, Runpod, Novita AI and Massed Compute links are referral links — if you sign up through one we may earn a commission at no cost to you. Lium links are plain links with no referral. That never changes the ranking: every price on this page is the live figure from the provider's API, cheapest first. Sponsored placements are labelled. Provider names and logos are their owners' trademarks, used to identify the services compared; no partnership or endorsement is implied. Prices are pulled hourly; VRAM fit is an estimate from each model's config, not a benchmark.