- Services
Every open model in our directory, ranked by how well it runs on NVIDIA H100 SXM5 80GB (80GB) at 4-bit. Grades weigh whether the model fits in memory and how fast it should run.
NVIDIA H100 SXM5 80GB has 80 GB of VRAM, enough to run 256 of the 297 open models we track at 4-bit. The largest that still fits well is Kimi K2 Instruct (1000B), and the top-rated is Apertus 70B at about 62 tok/s.
256 of 297 models run comfortably on NVIDIA H100 SXM5 80GB.
See full NVIDIA H100 SXM5 80GB specs and pricingThe top models this device can run at 4-bit, ranked by fit and speed.
| Model | Grade | Speed | VRAM |
|---|---|---|---|
| Apertus 70BEPFL, ETH Zurich, CSCS | SS | 62.1 tok/s | 43.4 GB |
| Llama 2 70B ChatMeta | SS | 62.1 tok/s | 43.4 GB |
| Llama 3 70B InstructMeta | SS | 59.0 tok/s | 45.7 GB |
| Mixtral 8x22B InstructMistral AI | SS | 61.9 tok/s | 43.6 GB |
| Hy3Tencent | SS | 48.4 tok/s | 55.7 GB |
| GLM-4.5Z.ai | SS | 52.0 tok/s | 51.8 GB |
| GLM-4.7Z.ai | SS | 51.3 tok/s | 52.6 GB |
| Qwen 3.5 OmniAlibaba | SS | 59.7 tok/s | 45.2 GB |
| Qwen3.5-397B-A17BAlibaba | SS | 58.6 tok/s | 46.0 GB |
| DeepSeek-V3DeepSeek | SS | 45.1 tok/s | 59.8 GB |
| DeepSeek-R1DeepSeek | SS | 45.1 tok/s | 59.8 GB |
| DeepSeek-V3.1DeepSeek | SS | 45.1 tok/s | 59.8 GB |
Apertus 70BEPFL, ETH Zurich, CSCS | 70B | SS | 62.1 tok/s | 43.4 GB | |
Llama 2 70B ChatMeta | 70B | SS | 62.1 tok/s | 43.4 GB | |
| 70B | SS | 59.0 tok/s | 45.7 GB | ||
Mixtral 8x22B InstructMistral AI | 141B(39B active) | SS | 61.9 tok/s | 43.6 GB | |
Hy3Tencent | 295B(21B active) | SS | 48.4 tok/s | 55.7 GB | |
GLM-4.5Z.ai | 355B(32B active) | SS | 52.0 tok/s | 51.8 GB | |
GLM-4.7Z.ai | 358B(32B active) | SS | 51.3 tok/s | 52.6 GB | |
Qwen 3.5 OmniAlibaba | 397B(17B active) | SS | 59.7 tok/s | 45.2 GB | |
Ad Tools for AI BuildersAll the hardware and tools you need to run private AI at your office or at home.Shop Now | |||||
Qwen3.5-397B-A17BAlibaba | 397B(17B active) | SS | 58.6 tok/s | 46.0 GB | |
DeepSeek-V3DeepSeek | 671B(37B active) | SS | 45.1 tok/s | 59.8 GB | |
DeepSeek-R1DeepSeek | 671B(37B active) | SS | 45.1 tok/s | 59.8 GB | |
DeepSeek-V3.1DeepSeek | 671B(37B active) | SS | 45.1 tok/s | 59.8 GB | |
DeepSeek-V3.2DeepSeek | 685B(37B active) | SS | 45.1 tok/s | 59.8 GB | |
Kimi K2 InstructMoonshot AI | 1000B(32B active) | SS | 52.0 tok/s | 51.8 GB | |
Qwen3-235B-A22BAlibaba | 235B(22B active) | SS | 74.2 tok/s | 36.3 GB | |
Gemma 3 27B ITGoogle | 27B | SS | 61.6 tok/s | 43.8 GB | |
Ad Vast.aiAffordable on-demand GPU rentals for training and inference. Pick from thousands of hosts.Rent a GPU | |||||
Muse Glimmer-30BMeta | 30B | SS | 54.6 tok/s | 49.4 GB | |
Qwen3-32BAlibaba | 32.8B | SS | 50.0 tok/s | 53.9 GB | |
Mistral Small 3 24BMistral AI | 24B | SS | 69.2 tok/s | 39.0 GB | |
LLaMA 65BMeta | 65B | SS | 68.7 tok/s | 39.3 GB | |
Qwen3.5-122B-A10BAlibaba | 122B(10B active) | SS | 98.9 tok/s | 27.3 GB | |
KolibriAleph Alpha | 78B(3B active) | SS | 102.8 tok/s | 26.2 GB | |
minimax-m2.5MiniMax | 230B(10B active) | SS | 118.8 tok/s | 22.7 GB | |
Gemma 4 12BGoogle | 12B | SS | 84.2 tok/s | 32.0 GB | |
| AdTools for AI BuildersAll the hardware and tools you need to run private AI at your office or at home.Shop Now | |||||
Gemma 4 12B Coderyuxinlu1 | 12B | SS | 84.2 tok/s | 32.0 GB | |
Model Compatibility
Use the calculator to weigh this device against any other, model by model, with speed and fit scores.
Straight answers to the questions we hear most often.
Can't find what you're looking for? Book a Discovery Call
Apertus 70B is the highest-rated model that runs on NVIDIA H100 SXM5 80GB at 4-bit, at about 62 tok/s. In total, NVIDIA H100 SXM5 80GB can run 256 of the 297 open models we track.
We size each model at 4-bit (Q4_K_M), add the KV cache for its context length and a fixed runtime overhead, then compare that to this device VRAM. Each model is graded on how comfortably it fits and how fast it should run on this card.
It is an estimate of decode speed in tokens per second at 4-bit, based on this device memory bandwidth and each model size. Real speed depends on your runtime, batch size, and context length, so use it to compare models, not as a hard promise.
Lower-bit quantization shrinks a model so more of them fit. These grades already assume 4-bit, which is the common local default. Going below 4-bit can squeeze a larger model in at some quality cost; going above needs more VRAM and may not fit.
Mixture-of-experts models only activate a fraction of their parameters at a time, so their runtime memory is closer to the active size than the headline size. We score fit on the active parameters, which is why some very large models still fit.
