AI models that fit on a 96 GB GPU
Open models that fit in 96 GB of VRAM: 17 as published, 17 at FP8 and 0 at INT4. Each model is listed on the smallest tier it fits at that precision, so anything smaller is on the 80 GB page or below.
GPUs with 96 GB
Cards whose datasheet VRAM puts them in this tier, up to the next tier at 141 GB. Prices are the lowest live data-center on-demand rate per GPU.
| GPU | VRAM | From per GPU hour |
|---|---|---|
| RTX PRO 6000 | 96 GB | $1.98/hr |
| RTX PRO 6000 SE | 96 GB | No live offer |
| RTX PRO 6000 WS | 96 GB | No live offer |
Fits as published
Models whose published weights, plus the flat overhead, fit this much memory with no quantization.
Text models
12 models.
| Model | Parameters | Published as | VRAM needed |
|---|---|---|---|
| Qwen3-Coder-Next-FP8 | 79.7B | F8_E4M3 | 89.0 GB |
| Ornith-1.5-35B-A3B | 36.0B | BF16 | 80.4 GB |
| Qwen3-Next-80B-A3B-Instruct-FP8 | 81.3B | F8_E4M3 | 90.9 GB |
| Phi-3.5-MoE-instruct | 41.9B | BF16 | 93.6 GB |
| Qwen3-Coder-Next-FP8 | 79.7B | F8_E4M3 | 89.0 GB |
| Qwen3-Coder-Next-FP8-dynamic | 79.8B | F8_E4M3 | 89.1 GB |
| Seed-OSS-36B-Instruct | 36.2B | BF16 | 80.8 GB |
| K2-Horizon-MoVA-36B-A4B | 37.4B | BF16 | 83.7 GB |
| falcon-40b | 41.8B | BF16 | 93.5 GB |
| Kolibri-1 | 78.1B | F8_E4M3 | 87.3 GB |
| Karnak-40B-v1.0 | 40.7B | BF16 | 90.9 GB |
| Apodex-1.1-mini | 36.0B | BF16 | 80.4 GB |
Vision-language models
4 models.
| Model | Parameters | Published as | VRAM needed |
|---|---|---|---|
| Qwen3.6-35B-A3B | 36.0B | BF16 | 80.4 GB |
| Qwen3.5-35B-A3B | 36.0B | BF16 | 80.4 GB |
| Qwen3.5-35B-A3B-Base | 36.0B | BF16 | 80.4 GB |
| fibo-scene-analyzer | 36.0B | BF16 | 80.4 GB |
Video generation models
1 model.
| Model | Parameters | Published as | VRAM needed |
|---|---|---|---|
| lingbot-world-fast | 18.5B | F32 | 82.9 GB |
Fits at FP8
Models that fit only after quantizing the weights to 8 bits (1 byte per parameter). Needs an FP8 checkpoint or an engine that quantizes on load, and a GPU with FP8 support.
Text models
15 models.
| Model | Parameters | Published as | VRAM needed |
|---|---|---|---|
| Qwen-72B | 72.3B | BF16 | 80.8 GB |
| Qwen3-Coder-Next | 79.7B | BF16 | 89.0 GB |
| Qwen2.5-72B-Instruct | 72.7B | BF16 | 81.3 GB |
| Qwen3-Next-80B-A3B-Instruct | 81.3B | BF16 | 90.9 GB |
| Hunyuan-A13B-Instruct | 80.4B | BF16 | 89.8 GB |
| Qwen2.5-72B | 72.7B | BF16 | 81.3 GB |
| Qwen3-Next-80B-A3B-Thinking | 81.3B | BF16 | 90.9 GB |
| Qwen2-72B-Instruct | 72.7B | BF16 | 81.3 GB |
| Le_Triomphant-ECE-TW3 | 72.3B | BF16 | 80.8 GB |
| TW3-JRGL-v2 | 72.3B | BF16 | 80.8 GB |
| Qwen2-72B | 72.7B | BF16 | 81.3 GB |
| Qwen1.5-72B-Chat | 72.3B | BF16 | 80.8 GB |
| Qwen1.5-72B | 72.3B | BF16 | 80.8 GB |
| AliceAI-Foundation-80B-A3B-Base | 81.3B | BF16 | 90.8 GB |
| Kolibri-1-BF16 | 78.1B | BF16 | 87.3 GB |
Vision-language models
2 models.
| Model | Parameters | Published as | VRAM needed |
|---|---|---|---|
| Qwen2.5-VL-72B-Instruct | 73.4B | BF16 | 82.0 GB |
| InternVL3-78B | 78.4B | BF16 | 87.6 GB |
Fits at INT4
Models that fit only after quantizing to 4 bits (0.5 byte per parameter). Needs a quantized checkpoint actually published for the model; check its Hugging Face page before relying on a row.
No catalog model has its smallest fit at this tier.
How these numbers are computed
Required VRAM is the weight size at each precision times a flat 1.2 overhead, the same figure every model page shows. It does not include a long context: the KV cache grows with every token, so a model near the top of a tier can need the next one at long context. Read how much VRAM you need for LLMs for the method, the VRAM and quantization glossary entries for the terms, and the VRAM calculator to size a model that is not listed.
Looking for a model by job rather than by memory? Start with coding, reasoning, chat and assistants, vision-language or see the full models directory.
