Skip to content

What can the 4× H200 NVL run?

What fits on the 4× H200 NVL
Text models: 59Visual models: 0Strongest first
Memory
564 GB
Bandwidth
4,800 GB/s each
Maker
NVIDIA

Open the calculator on this machineH200 NVL in our inventory

Text models that run on it

Text models that run on the 4× H200 NVL, best first
ModelBest fitContextSpeedStrongest at
MiMo V2.6 ProFP4 + FP8112K71 tok/sChat, Coding, Agents, Docs
DeepSeek V4.1 FlashFP4 + FP81M135 tok/sChat, Coding, Agents, Docs
GLM-5.35-bit352K76 tok/sChat, Coding, Agents, Docs
GLM-5.3 FlashFP81M122 tok/sChat, Coding, Agents, Docs
Qwen3.8 2.4T A95B1.6-bit256K87 tok/sChat, Coding, Agents
Qwen3.8 Flash-Next8-bit256K208 tok/sChat, Coding, Agents
Muse Spark Open 816B A42B4-bit1M89 tok/sChat, Docs
MiMo V2.6 FlashFP4 + FP81M164 tok/sCoding, Agents
MiniMax M38-bit928K92 tok/sDocs
DeepSeek V4 FlashFP4 + FP81M122 tok/sDocs
DeepSeek V4 Pro1.6-bit1M150 tok/sChat, Coding, Agents, Docs
Kimi K2.63-bit256K120 tok/sChat, Coding, Agents, Docs
Hy38-bit256K87 tok/sChat, Coding, Agents, Docs
Nemotron 3 Ultra 550B A55B6-bit256K50 tok/sChat, Coding, Docs
Qwen3.5 397B A17B8-bit256K111 tok/sAgents, Docs
MiniMax M2.7FP8200K149 tok/sCoding, Agents
DeepSeek V3.26-bit160K79 tok/sChat, Coding, Agents, Docs
Inkling4-bit4K88 tok/sText and images
DeepSeek R16-bit144K74 tok/sNot rated
DeepSeek V36-bit144K74 tok/sNot rated
GLM 55-bit198K73 tok/sNot rated
Inkling Small8-bit4K163 tok/sText and images
MiniMax M2.5FP8192K139 tok/sNot rated
GLM 5.15-bit198K73 tok/sNot rated
Laguna M.18-bit256K86 tok/sNot rated
Qwen3 VL 235B A22B Instruct8-bit256K79 tok/sText and images
GLM 5.25-bit352K74 tok/sNot rated
Qwen3 235B A22B8-bit40K79 tok/sNot rated
Kimi K2 Instruct3-bit128K103 tok/sNot rated
Ornith 1.5 397B8-bit256K88 tok/sNot rated
Ornith 1.0 397B8-bit256K108 tok/sNot rated
Step 3.5 Flash8-bit256K176 tok/sNot rated
DeepSeek V3 03246-bit144K74 tok/sNot rated
DeepSeek R1 05286-bit144K74 tok/sNot rated
GLM 4.58-bit128K60 tok/sNot rated
MiniMax M2FP8192K139 tok/sNot rated
MiMo V2.5FP81M123 tok/sNot rated
GLM 5.3 Flash BF168-bit1M117 tok/sText and images
Qwen3 VL 235B A22B Thinking8-bit256K79 tok/sText and images
Qwen3 235B A22B Instruct 25078-bit256K79 tok/sNot rated
Hy3 Preview8-bit256K88 tok/sNot rated
Qwen3 Coder 480B A35B Instruct8-bit232K68 tok/sCoding
GLM 4.78-bit198K60 tok/sNot rated
GLM 5.3 BF165-bit352K74 tok/sNot rated
GLM 4.68-bit198K61 tok/sNot rated
DeepSeek V2 Chat8-bit160K97 tok/sNot rated
MiMo V2 FlashFP8256K128 tok/sNot rated
Le Triomphant ECE TW38-bit32K40 tok/sNot rated
TW3 JRGL V28-bit32K40 tok/sNot rated
Command A Plus 05 2026 Bf168-bit195K100 tok/sText and images
K EXAONE 2.0 750B A37B5-bit256K80 tok/sNot rated
MiniMax M2.1FP8192K139 tok/sNot rated
Beam 1.0 501B A23B8-bit1M96 tok/sNot rated
Hy4 Preview5-bit168K66 tok/sNot rated
Qwen1.5 72B8-bit32K40 tok/sNot rated
WizardLM 2 8x22B8-bit64K64 tok/sNot rated
Intern S18-bit64K69 tok/sText and images
Step 3.7 Flash8-bit256K175 tok/sText and images
K2 Horizon 375B A23B8-bit512K91 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%.

Only just too big

These need up to a quarter more memory than the 4× H200 NVL has. Spilling into system RAM or a lower precision can get them running, slowly.

Questions

What is the best LLM for the 4× H200 NVL?

Of the open models we track, the strongest that run well on the 4× H200 NVL are MiMo V2.6 Pro, DeepSeek V4.1 Flash, GLM-5.3. 59 text models fit it in all.

Calculated from published specs, not benchmarked. All machines