VRAMNeeded Calculator

GPU VRAM lookup

Search a graphics card, data-center GPU or Mac to see how much memory it has, which language models fit on it with an 8K-token conversation, and what it costs to rent one by the hour. Filter by minimum memory to compare cards you could buy or rent.

GPU or MacMemoryLLMs that fit (8K context)Largest that fitsCheapest rental (1 GPU)More
RTX 3060NVIDIA GeForce12 GB
10 of 30

Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen2.5 14B Instruct4-bit (Q4_K_M), 10.6 GB$0.05/hrVast.ai, Marketplace, verified on-demandFull list
RTX 3090NVIDIA GeForce24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$0.16/hrVast.ai, Marketplace, verified on-demandFull list
RTX 4060NVIDIA GeForce8 GB
8 of 30

DeepSeek-R1-0528-Qwen3-8B (4-bit), Llama 3.1 8B Instruct (4-bit), Nemotron 3.5 Bolt E4B (4-bit), Qwen2.5 7B Instruct (4-bit), Mistral 7B Instruct v0.3 (4-bit), Spark-X2.6 4B (8-bit), Llama 3.2 3B Instruct (8-bit), LFM3 1.7B (16-bit)

DeepSeek-R1-0528-Qwen3-8B4-bit (Q4_K_M), 6.2 GBnot trackedFull list
RTX 4060 Ti 8 GBNVIDIA GeForce8 GB
8 of 30

DeepSeek-R1-0528-Qwen3-8B (4-bit), Llama 3.1 8B Instruct (4-bit), Nemotron 3.5 Bolt E4B (4-bit), Qwen2.5 7B Instruct (4-bit), Mistral 7B Instruct v0.3 (4-bit), Spark-X2.6 4B (8-bit), Llama 3.2 3B Instruct (8-bit), LFM3 1.7B (16-bit)

DeepSeek-R1-0528-Qwen3-8B4-bit (Q4_K_M), 6.2 GBnot trackedFull list
RTX 4060 Ti 16 GBNVIDIA GeForce16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
RTX 4070NVIDIA GeForce12 GB
10 of 30

Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen2.5 14B Instruct4-bit (Q4_K_M), 10.6 GBnot trackedFull list
RTX 4080NVIDIA GeForce16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
RTX 4090NVIDIA GeForce24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$0.31/hrVast.ai, Marketplace, verified on-demandFull list
RTX 5060NVIDIA GeForce8 GB
8 of 30

DeepSeek-R1-0528-Qwen3-8B (4-bit), Llama 3.1 8B Instruct (4-bit), Nemotron 3.5 Bolt E4B (4-bit), Qwen2.5 7B Instruct (4-bit), Mistral 7B Instruct v0.3 (4-bit), Spark-X2.6 4B (8-bit), Llama 3.2 3B Instruct (8-bit), LFM3 1.7B (16-bit)

DeepSeek-R1-0528-Qwen3-8B4-bit (Q4_K_M), 6.2 GBnot trackedFull list
RTX 5070NVIDIA GeForce12 GB
10 of 30

Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen2.5 14B Instruct4-bit (Q4_K_M), 10.6 GBnot trackedFull list
RTX 5070 TiNVIDIA GeForce16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
RTX 5080NVIDIA GeForce16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
RTX 5090NVIDIA GeForce32 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.50/hrVast.ai, Marketplace, verified on-demandFull list
RX 7900 XTXAMD Radeon24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GBnot trackedFull list
Arc A770 16 GBIntel Arc16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
Mac, 16 GBApple Mac (unified memory)16 GB
11 of 30

gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (4-bit), Qwen3 14B (4-bit), DeepSeek-R1-0528-Qwen3-8B (8-bit), Llama 3.1 8B Instruct (8-bit), Nemotron 3.5 Bolt E4B (8-bit), Qwen2.5 7B Instruct (8-bit), Mistral 7B Instruct v0.3 (8-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-20b16-bit (FP16/BF16), 14.0 GBnot trackedFull list
Mac, 24 GBApple Mac (unified memory)24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GBnot trackedFull list
Mac, 32 GBApple Mac (unified memory)32 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GBnot trackedFull list
Mac, 36 GBApple Mac (unified memory)36 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (4-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GBnot trackedFull list
Mac, 48 GBApple Mac (unified memory)48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GBnot trackedFull list
Mac, 64 GBApple Mac (unified memory)64 GB
24 of 30

Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Llama 3.1 70B Instruct4-bit (Q4_K_M), 45.2 GBnot trackedFull list
Mac, 96 GBApple Mac (unified memory)96 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GBnot trackedFull list
Mac, 128 GBApple Mac (unified memory)128 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GBnot trackedFull list
Mac Studio, 256 GBApple Mac (unified memory)256 GB
27 of 30

Qwen3.8 Flash Next (8-bit), Ornith-1.5-122B (8-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (16-bit), Llama 3.3 70B Instruct (16-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next8-bit (Q8_0), 188.5 GBnot trackedFull list
Mac Studio, 512 GBApple Mac (unified memory)512 GB
27 of 30

Qwen3.8 Flash Next (16-bit), Ornith-1.5-122B (16-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (16-bit), Llama 3.3 70B Instruct (16-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next16-bit (FP16/BF16), 354.2 GBnot trackedFull list
NVIDIA A100 40 GBData center40 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$1.99/hrLambda, On-demand instance, A100 PCIeFull list
NVIDIA A100 80 GBData center80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$0.87/hrVast.ai, Marketplace, verified on-demand, A100 PCIeFull list
NVIDIA H100 80 GBData center80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$1.99/hrRunpod, Community Cloud, H100 PCIeFull list
NVIDIA RTX PRO 6000 Blackwell 96 GBData center96 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GBnot trackedFull list
NVIDIA DGX Spark 128 GBData center128 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GBnot trackedFull list
NVIDIA H200 141 GBData center141 GB
27 of 30

Qwen3.8 Flash Next (4-bit), Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next4-bit (Q4_K_M), 119.8 GB$3.59/hrRunpod, Community Cloud, H200Full list
A10 24 GBCloud GPU24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$1.29/hrLambda, On-demand instanceFull list
L4 24 GBCloud GPU24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$0.44/hrRunpod, Community CloudFull list
Quadro RTX 6000 24 GBCloud GPU24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$0.69/hrLambda, On-demand instanceFull list
RTX A5000 24 GBCloud GPU24 GB
15 of 30

Gemma 4 31B (4-bit), Qwen3.8 27B (4-bit), Gemma 4 26B A4B (4-bit), LFM2.5-24B-A2B (4-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (8-bit), Qwen3 14B (8-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Gemma 4 31B4-bit (Q4_K_M), 19.6 GB$0.16/hrRunpod, Community CloudFull list
A100 PCIe 40 GBCloud GPU40 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$1.99/hrLambda, On-demand instanceFull list
A100 SXM 40 GBCloud GPU40 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (4-bit), Qwen3.6 35B-A3B (4-bit), Qwen3.8-35B-A3B (Empero Distill) (4-bit), Nanbeige4.3 33B (4-bit), Qwen2.5 32B Instruct (4-bit), Qwen3 32B (4-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$1.99/hrLambda, On-demand instanceFull list
A40 48 GBCloud GPU48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.35/hrRunpod, Community CloudFull list
L40 48 GBCloud GPU48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.69/hrRunpod, Community CloudFull list
L40S 48 GBCloud GPU48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.53/hrVast.ai, Marketplace, verified on-demandFull list
RTX 6000 Ada 48 GBCloud GPU48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.74/hrRunpod, Community CloudFull list
RTX A6000 48 GBCloud GPU48 GB
22 of 30

Mixtral 8x7B Instruct v0.1 (4-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (8-bit), Qwen3.8-35B-A3B (Empero Distill) (8-bit), Nanbeige4.3 33B (8-bit), Qwen2.5 32B Instruct (8-bit), Qwen3 32B (8-bit), Gemma 4 31B (8-bit), Qwen3.8 27B (8-bit), Gemma 4 26B A4B (8-bit), LFM2.5-24B-A2B (8-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Mixtral 8x7B Instruct v0.14-bit (Q4_K_M), 27.5 GB$0.33/hrRunpod, Community CloudFull list
A100 PCIe 80 GBCloud GPU80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$0.87/hrVast.ai, Marketplace, verified on-demandFull list
A100 SXM 80 GBCloud GPU80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$1.07/hrVast.ai, Marketplace, verified on-demandFull list
H100 PCIe 80 GBCloud GPU80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$1.99/hrRunpod, Community CloudFull list
H100 SXM 80 GBCloud GPU80 GB
25 of 30

gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (4-bit), Llama 3.3 70B Instruct (4-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (8-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

gpt-oss-120b16-bit (FP16/BF16), 65.7 GB$2.14/hrVast.ai, Marketplace, verified on-demandFull list
H100 NVL 94 GBCloud GPU94 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GB$2.59/hrRunpod, Community CloudFull list
GH200 96 GBCloud GPU96 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GB$2.29/hrLambda, On-demand instanceFull list
RTX Pro 6000 96 GBCloud GPU96 GB
26 of 30

Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (8-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Ornith-1.5-122B4-bit (Q4_K_M), 79.9 GB$1.69/hrRunpod, Community CloudFull list
H200 141 GBCloud GPU141 GB
27 of 30

Qwen3.8 Flash Next (4-bit), Ornith-1.5-122B (4-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (8-bit), Llama 3.3 70B Instruct (8-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next4-bit (Q4_K_M), 119.8 GB$3.59/hrRunpod, Community CloudFull list
B200 180 GBCloud GPU180 GB
27 of 30

Qwen3.8 Flash Next (4-bit), Ornith-1.5-122B (8-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (16-bit), Llama 3.3 70B Instruct (16-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next4-bit (Q4_K_M), 119.8 GB$5.98/hrRunpod, Community CloudFull list
B200 SXM6 180 GBCloud GPU180 GB
27 of 30

Qwen3.8 Flash Next (4-bit), Ornith-1.5-122B (8-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (16-bit), Llama 3.3 70B Instruct (16-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next4-bit (Q4_K_M), 119.8 GB$6.99/hrLambda, On-demand instanceFull list
B300 288 GBCloud GPU288 GB
27 of 30

Qwen3.8 Flash Next (8-bit), Ornith-1.5-122B (16-bit), gpt-oss-120b (16-bit), Llama 3.1 70B Instruct (16-bit), Llama 3.3 70B Instruct (16-bit), Mixtral 8x7B Instruct v0.1 (16-bit), Recursive Flint (16-bit), Qwen3.6 35B-A3B (16-bit), Qwen3.8-35B-A3B (Empero Distill) (16-bit), Nanbeige4.3 33B (16-bit), Qwen2.5 32B Instruct (16-bit), Qwen3 32B (16-bit), Gemma 4 31B (16-bit), Qwen3.8 27B (16-bit), Gemma 4 26B A4B (16-bit), LFM2.5-24B-A2B (16-bit), gpt-oss-20b (16-bit), Qwen2.5 14B Instruct (16-bit), Qwen3 14B (16-bit), DeepSeek-R1-0528-Qwen3-8B (16-bit), Llama 3.1 8B Instruct (16-bit), Nemotron 3.5 Bolt E4B (16-bit), Qwen2.5 7B Instruct (16-bit), Mistral 7B Instruct v0.3 (16-bit), Spark-X2.6 4B (16-bit), Llama 3.2 3B Instruct (16-bit), LFM3 1.7B (16-bit)

Qwen3.8 Flash Next8-bit (Q8_0), 188.5 GB$6.94/hrRunpod, Community CloudFull list

"Fit" uses the same rule as the VRAM calculator and what can I run: the highest precision (16-bit, 8-bit or 4-bit Q4_K_M) whose Hugging Face file size plus an FP16 KV cache for 8,192 tokens stays under 90% of the card's memory. 30 language models, model data checked 2026-10-30. Rental price = cheapest listed single-GPU on-demand rate in our GPU price table (Runpod, Lambda and Vast.ai, read 2026-10-31); prices change, so check the provider before you rent. Mac rows use total unified memory; macOS keeps part of it for the system, so treat Mac fits as optimistic.

How to use it

  • Have a card already? Find it, open the "LLMs that fit" count to see every model, then use Full list for the per-precision table.
  • Shopping for a card? Set a minimum memory and compare the "Largest that fits" column. The buying guides cover specific cards at 12, 16, 24 and 32 GB.
  • Need more memory than you own? The "Cheapest rental" column links to our GPU price table. For the same card on Vast.ai and Runpod side by side, see Vast.ai vs Runpod cost per hour.
  • Longer context? An 8K conversation is the default here. Long contexts need much more memory for the KV cache: check your exact setup in the VRAM calculator.

Where the numbers come from

  • Memory per card: the same list as the homepage calculator (NVIDIA, AMD, Intel, Apple and data-center GPUs) plus the cloud GPUs in our price table.
  • Model sizes: real Hugging Face download sizes for each model (Q4_K_M, Q8_0 and BF16 files), plus the KV cache worked out from each model's config. The same data is in the comparison table and as JSON.
  • Rule of thumb: "fits" means under 90% of memory, leaving room for the runtime. That line is our own rule, not a vendor figure.

Last checked: