AI Compatibility Tier List
NVIDIA H20 96GB
374 AI models ranked by how well they run on your 96 GB VRAM hardware. Best match: Qwen 3.5 122B A10B (score: 96).
374
Models tested
330
Can run
96 GB
VRAM
130
Best tok/s
Tier Distribution
S 31
A 57
B 56
C 186
F 44
S: 31A: 57B: 56C: 186F: 44
S
S Tier -- Perfect Fit (31 models)
These models run flawlessly with headroom to spare.
Qwen 3.5 122B A10B122B87.4 GB130 tok/s73K ctx
moe
Qwen3-Coder-Next80B60.8 GB219 tok/s256K ctx
moe
Qwen 2.5 VL 72B72B59.3 GB80 tok/s33K ctx
dense
Devstral 2 123B Instruct123B90.9 GB47 tok/s31K ctx
dense
Mistral Small 4 119B119B88.5 GB141 tok/s38K ctx
moe
Command A 111B111B82.1 GB52 tok/s73K ctx
dense
GPT-OSS 120B117B86.8 GB49 tok/s46K ctx
dense
Qwen 3.6 35B A3B35B36.0 GB412 tok/s250K ctx
moe
Qwen3-Coder 30B A3B Instruct30.5B30.6 GB490 tok/s256K ctx
moe
Pixtral Large 124B124B91.5 GB47 tok/s29K ctx
dense
Qwen 3.5 27B27B30.1 GB213 tok/s131K ctx
dense
Qwen3-VL 30B A3B Instruct30B30.3 GB507 tok/s256K ctx
moe
Qwen 3.5 35B A3B35B33.3 GB448 tok/s131K ctx
moe
Qwen 3.6 27B27B27.9 GB132 tok/s262K ctx
dense
Qwen 3 32B32B33.9 GB181 tok/s131K ctx
dense
Leanstral 119B A6B119B91.9 GB130 tok/s24K ctx
moe
Magistral Small 250724B27.6 GB238 tok/s131K ctx
dense
Devstral Small 2 24B Instruct24B27.6 GB238 tok/s256K ctx
dense
Qwen 3 30B A3B30.5B30.6 GB490 tok/s131K ctx
moe
Nemotron 3 Nano 30B30B31.2 GB190 tok/s131K ctx
dense
Llama 3.3 70B70B58.1 GB83 tok/s128K ctx
dense
Gemma 4 31B30.7B43.9 GB113 tok/s73K ctx
dense
Devstral Small 1.124B27.6 GB238 tok/s131K ctx
dense
Qwen 3.5 9B9B18.2 GB126 tok/s131K ctx
dense
Qwen 3 14B14B21.5 GB196 tok/s131K ctx
dense
Nemotron Cascade 2 30B A3B30B31.7 GB501 tok/s262K ctx
moe
Phi-4-reasoning-plus 14B14.7B22.5 GB206 tok/s33K ctx
dense
GPT-OSS 20B21B25.8 GB622 tok/s128K ctx
moe
Qwen 3 8B8B17.6 GB112 tok/s131K ctx
dense
Qwen 2.5 72B72B59.3 GB80 tok/s131K ctx
dense
Llama 3.1 70B70B58.1 GB83 tok/s128K ctx
dense
A
A Tier -- Great Fit (57 models)
Excellent performance with comfortable memory margins.
EXAONE 4.0 32B32B33.9 GB179 tok/s131K ctx
dense
Qwen 3.5 4B4B15.1 GB56 tok/s131K ctx
dense
Gemma 4 26B A4B25.2B29.5 GB526 tok/s256K ctx
moe
Kimi Linear 48B A3B48B40.7 GB111 tok/s970K ctx
linear
Qwen 2.5 32B32B33.9 GB179 tok/s131K ctx
dense
Gemma 3 27B27B38.2 GB128 tok/s98K ctx
dense
Codestral 2 25.0822B26.4 GB232 tok/s256K ctx
dense
Mistral Small 3.2 24B24B27.6 GB238 tok/s131K ctx
vision
OLMo 2 32B32B33.9 GB179 tok/s4K ctx
dense
Llama 4 Scout 17B 16E109B79.9 GB124 tok/s104K ctx
moe
Ministral 3 14B14B21.5 GB196 tok/s262K ctx
multimodal
Granite 4.1 30B30B32.7 GB190 tok/s131K ctx
dense
LFM2 24B24B27.6 GB238 tok/s131K ctx
dense
DeepSeek R1 Distill 70B70B58.1 GB83 tok/s131K ctx
dense
Nemotron Nano 8B8B17.3 GB112 tok/s131K ctx
dense
CogVLM2 19B19B24.5 GB266 tok/s8K ctx
dense
Phi-4 Mini Reasoning 4B3.8B14.3 GB53 tok/s131K ctx
dense
Mistral Small 24B24B27.6 GB238 tok/s33K ctx
dense
Mistral Small 3.1 24B24B27.6 GB238 tok/s131K ctx
dense
Phi-4 14B14B22.1 GB196 tok/s16K ctx
dense
InternVL2 8B8B17.3 GB112 tok/s8K ctx
dense
Qwen 2.5 14B14B22.0 GB196 tok/s131K ctx
dense
Granite Code 20B20B25.9 GB280 tok/s8K ctx
dense
Qwen 2.5 Coder 32B32B33.9 GB179 tok/s131K ctx
dense
Gemma 3 12B12B22.7 GB168 tok/s131K ctx
dense
Granite Code 34B34B34.9 GB169 tok/s8K ctx
dense
Ministral 3 8B8B17.6 GB112 tok/s262K ctx
multimodal
MiniCPM-V 2.6 8B8B17.3 GB112 tok/s2K ctx
dense
Nemotron 70B70B58.1 GB83 tok/s131K ctx
dense
Command R 35B35B34.3 GB165 tok/s131K ctx
dense
DeepSeek Coder V2 16B16B23.6 GB790 tok/s131K ctx
moe
Nemotron Nano 9B v29B18.4 GB126 tok/s131K ctx
dense
SQLCoder 7B7B16.7 GB98 tok/s8K ctx
dense
Qwen 3 4B4B15.1 GB56 tok/s33K ctx
dense
StarCoder 15B15B35.9 GB210 tok/s8K ctx
dense
Qwen 2.5 VL 7B7B15.6 GB98 tok/s33K ctx
dense
MPT-30B-Instruct30B55.5 GB153 tok/s8K ctx
dense
DeepSeek R1 Distill 32B32B33.9 GB179 tok/s33K ctx
dense
Jina Embeddings v30.57B13.6 GB8 tok/s8K ctx
dense
CodeLlama 13B Instruct13B30.6 GB182 tok/s16K ctx
dense
OLMo 2 13B13B20.9 GB182 tok/s33K ctx
dense
Magistral 7B7B16.7 GB98 tok/s8K ctx
dense
CodeGeeX 4 9B9B16.6 GB126 tok/s131K ctx
dense
Gemma 4 E4B8B16.7 GB112 tok/s128K ctx
dense
BGE M30.57B12.8 GB8 tok/s8K ctx
dense
Command R+ 104B104B77.4 GB56 tok/s103K ctx
dense
LLaVA 1.6 13B13B30.6 GB182 tok/s4K ctx
dense
Falcon 40B Instruct40B41.1 GB125 tok/s8K ctx
dense
DeepSeek R1 Distill 14B14B22.0 GB196 tok/s33K ctx
dense
Qwen 2.5 7B7B15.6 GB98 tok/s131K ctx
dense
Granite Code 8B8B17.3 GB112 tok/s8K ctx
dense
DevStral 7B7B16.7 GB98 tok/s8K ctx
dense
Codestral Mamba 7B7B15.3 GB98 tok/s262K ctx
state-space
CodeLlama 7B Instruct7B22.6 GB98 tok/s16K ctx
dense
StarCoder 7B7B22.1 GB98 tok/s8K ctx
dense
Granite 4.1 8B8B17.8 GB112 tok/s131K ctx
dense
Pixtral 12B12B20.3 GB168 tok/s131K ctx
dense
B
B Tier -- Good Fit (56 models)
Solid performance, may need quantization or reduced context.
Gemma 2 27B27B38.2 GB128 tok/s8K ctx
dense
Nous Dolphin 13B13B32.1 GB182 tok/s16K ctx
dense
mxbai Embed Large0.34B12.7 GB5 tok/s1K ctx
dense
WizardLM 13B13B30.6 GB182 tok/s8K ctx
dense
Vicuna 13B13B30.6 GB182 tok/s4K ctx
dense
Nous Hermes 1.09B28.2 GB126 tok/s16K ctx
dense
Snowflake Arctic Embed L0.34B12.7 GB5 tok/s1K ctx
dense
Qwen 2.5 Coder 3B3B14.5 GB42 tok/s131K ctx
dense
Qwen 2.5 Math 72B72B59.3 GB80 tok/s4K ctx
dense
InternLM 7B7B22.6 GB98 tok/s8K ctx
dense
InternLM Chat 7B7B22.6 GB98 tok/s8K ctx
dense
Gemma 4 E2B5.1B14.1 GB71 tok/s128K ctx
dense
GLM-4 9B9B16.6 GB126 tok/s128K ctx
dense
BGE Large EN v1.50.34B12.7 GB5 tok/s1K ctx
dense
Llama 3.1 8B8B17.3 GB112 tok/s128K ctx
dense
Mixtral 8x7B47B41.1 GB233 tok/s33K ctx
moe
Solar 7B7B17.7 GB98 tok/s8K ctx
dense
WizardMath 7B7B16.7 GB98 tok/s4K ctx
dense
OLMo 2 7B7B16.7 GB98 tok/s4K ctx
dense
Mixtral 8x22B141B99.9 GB63 tok/s4K ctx
moe
Nomic Embed Text v1.50.14B11.3 GB2 tok/s8K ctx
dense
Qwen 2.5 Coder 7B7B15.6 GB98 tok/s131K ctx
dense
Baichuan 13B13B32.1 GB182 tok/s8K ctx
dense
Gemma 3 4B4B15.0 GB56 tok/s128K ctx
dense
LLaVA 1.5 7B7B22.6 GB98 tok/s4K ctx
dense
Ministral 3 3B3B13.1 GB42 tok/s262K ctx
multimodal
Cerebras-GPT 13B13B29.6 GB182 tok/s125K ctx
dense
DeepSeek LLM 67B67B57.2 GB86 tok/s4K ctx
dense
Phi 4 Mini 4B4B14.4 GB56 tok/s128K ctx
dense
Qwen 3.5 2B2B13.4 GB28 tok/s131K ctx
dense
MPT-7B-Instruct7B22.6 GB98 tok/s8K ctx
dense
Baichuan 7B7B22.6 GB98 tok/s8K ctx
dense
DeepSeek R1 Distill 7B7B15.6 GB98 tok/s33K ctx
dense
Yi 1.5 34B34B34.9 GB170 tok/s4K ctx
dense
DeepSeek R1 Distill 8B8B17.3 GB112 tok/s33K ctx
dense
Samantha 7B7B16.7 GB98 tok/s4K ctx
dense
Qwen 2.5 3B3B14.5 GB42 tok/s131K ctx
dense
Falcon 7B Instruct7B14.9 GB98 tok/s8K ctx
dense
Phi 3 Mini 3.8B3.8B18.7 GB53 tok/s128K ctx
dense
Qwen 2.5 Coder 14B14B22.0 GB196 tok/s131K ctx
dense
Gemma 2 9B9B21.1 GB126 tok/s8K ctx
dense
Llama 3.2 11B Vision11B19.2 GB154 tok/s16K ctx
vision
Qwen 3 1.7B1.7B13.2 GB24 tok/s33K ctx
dense
Phi 3.5 Mini 4B4B18.8 GB56 tok/s128K ctx
dense
All MiniLM L6 v20.02B10.8 GB2 tok/s0K ctx
dense
Granite 4.1 3B3B13.6 GB42 tok/s131K ctx
dense
InternLM 20B20B43.2 GB230 tok/s8K ctx
dense
Granite Code 3B3B14.8 GB42 tok/s8K ctx
dense
Mistral Nemo 12B12B20.3 GB168 tok/s128K ctx
dense
Qwen 2.5 Coder 1.5B1.5B11.8 GB21 tok/s33K ctx
dense
Yi Coder 9B9B17.5 GB126 tok/s131K ctx
dense
Codestral 22B22B26.4 GB247 tok/s33K ctx
dense
Mistral 7B Instruct v0.37B16.7 GB98 tok/s8K ctx
dense
Phi 3 Medium 14B14B22.1 GB196 tok/s128K ctx
dense
Ministral 8B8B17.6 GB112 tok/s131K ctx
dense
Llama 3.3 70B Instruct
B5570B61.4 GB76 tok/s83K ctx
dense
C
C Tier -- Tight Fit (186 models)
Runs with offloading or significant compromises.
Llama 3.2 3B3B14.0 GB42 tok/s128K ctx
dense
BaichuanMed OCR 72B i1
C5572B62.9 GB74 tok/s79K ctx
dense
Solar Open 69B REAP i1
C5569B60.7 GB77 tok/s86K ctx
dense
stabilityai japanese stablelm instruct beta 70b
C5570B61.4 GB76 tok/s83K ctx
dense
Aya Expanse 32B32B32.5 GB181 tok/s8K ctx
dense
Qwen3.5 122B A10B122B84.6 GB50 tok/s29K ctx
dense
Solar Open 100B
C52100B83.2 GB53 tok/s33K ctx
dense
Solar Open 100B i1
C52100B83.2 GB53 tok/s33K ctx
dense
Qwen3 48B A4B Savant Commander Distill 12X Closed Open Heretic Uncensored
C5248B45.4 GB111 tok/s160K ctx
dense
Yi 34B Chat34B34.9 GB170 tok/s200K ctx
dense
DeepSeek R1 1.5B1.5B11.8 GB21 tok/s33K ctx
dense
TinyLlama 1.1B1.1B11.5 GB15 tok/s4K ctx
dense
Granite 3.1 8B8B17.3 GB112 tok/s128K ctx
state-space
SmolLM3 3B3B14.3 GB42 tok/s128K ctx
dense
Qwen3.5 35B A3B35B36.0 GB152 tok/s250K ctx
dense
Yi 1.5 9B9B17.5 GB126 tok/s4K ctx
dense
Qwen3.5 35B A3B
C5035B36.0 GB152 tok/s250K ctx
dense
Hermes 4.3 36B
C5036B36.7 GB148 tok/s241K ctx
dense
Qwen 2.5 Math 7B7B15.6 GB98 tok/s4K ctx
dense
EXAONE 4.0 32B
C4932B33.8 GB166 tok/s282K ctx
dense
aya expanse 32b heretic MPOA i1
C4932B33.8 GB166 tok/s282K ctx
dense
Baichuan M2 32B Q4 K M
C4932B33.8 GB166 tok/s282K ctx
dense
baichuan inc Baichuan M2 32B
C4932B33.8 GB166 tok/s282K ctx
dense
Qwen3.5 27B27B30.1 GB197 tok/s349K ctx
dense
gemma 3 27b it27B30.1 GB197 tok/s349K ctx
dense
gemma 3 27b it
C4927B30.1 GB197 tok/s349K ctx
dense
StableLM 2 12B12B31.3 GB168 tok/s4K ctx
dense
StarCoder2 15B15B22.5 GB210 tok/s16K ctx
dense
Mistral Small 3.2 24B Instruct 250624B28.0 GB221 tok/s403K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
C4824B28.0 GB221 tok/s403K ctx
dense
Mistral Small 24B Instruct 2501
C4824B28.0 GB221 tok/s403K ctx
dense
mistral small 3.1 24b instruct 2503 hf
C4824B28.0 GB221 tok/s403K ctx
dense
cognitivecomputations Dolphin3.0 R1 Mistral 24B
C4824B28.0 GB221 tok/s403K ctx
dense
Dolphin Mistral GLM 4.7 Flash 24B Venice Edition Thinking Uncensored i1
C4824B28.0 GB221 tok/s403K ctx
dense
Qwen 2.5 1.5B1.5B11.8 GB21 tok/s131K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
C4824B28.0 GB221 tok/s403K ctx
dense
Codestral 22B v0.1
C4822B26.5 GB241 tok/s447K ctx
dense
Codestral 22B v0.1
C4822B26.5 GB241 tok/s447K ctx
dense
OpenChat 7B7B16.7 GB98 tok/s8K ctx
dense
Codestral 22B v0.1
C4822B26.5 GB241 tok/s447K ctx
dense
Vicuna 7B7B22.6 GB98 tok/s4K ctx
dense
Codestral 22B v0.1 IMat
C4722B26.5 GB241 tok/s447K ctx
dense
Aya Expanse 8B8B17.3 GB112 tok/s8K ctx
dense
Codestral 22B v0.1 i1
C4722B26.5 GB241 tok/s447K ctx
dense
Codestral 21B Pruned i1
C4721B25.8 GB253 tok/s473K ctx
dense
internlm2 5 20b chat
C4720B25.0 GB266 tok/s500K ctx
dense
internlm2 limarp chat 20b
C4720B25.0 GB266 tok/s500K ctx
dense
internlm2 math plus 20b i1
C4720B25.0 GB266 tok/s500K ctx
dense
Codestral RAG 19B Pruned i1
C4719B24.3 GB266 tok/s531K ctx
dense
DeepSeek R1 Distill Qwen 14B14B20.7 GB196 tok/s751K ctx
dense
Phi 4 reasoning vision 15B
C4715B21.4 GB210 tok/s695K ctx
dense
StarCoder2 15B
C4715B21.4 GB210 tok/s695K ctx
dense
Gemma 3 1B1B11.5 GB14 tok/s33K ctx
dense
OpenHermes 2.5 7B7B16.7 GB98 tok/s8K ctx
dense
Dolphin 2.9 8B8B17.3 GB112 tok/s33K ctx
dense
NousResearch Hermes 4 14B
C4614B20.7 GB196 tok/s751K ctx
dense
starcoder2 15b instruct v0.1
C4615B21.4 GB210 tok/s695K ctx
dense
starcoder2 15b i1
C4615B21.4 GB210 tok/s695K ctx
dense
starcoder2 15b instruct v0.1
C4615B21.4 GB210 tok/s695K ctx
dense
gemma 3 12b it
C4612B19.2 GB168 tok/s890K ctx
dense
HelpingAI 15B i1
C4615B21.4 GB210 tok/s695K ctx
dense
Zephyr 7B Beta7B16.7 GB98 tok/s33K ctx
dense
GGUF SOLARized GraniStral 14B 2102 YeAM HCT 32QKV
C4614B20.7 GB196 tok/s751K ctx
dense
internlm JanusCoder 14B
C4614B20.7 GB196 tok/s751K ctx
dense
GGUF SOLARized GraniStral 14B 1902 YeAM HCT
C4614B20.7 GB196 tok/s751K ctx
dense
Qwen3.5 9B9B17.0 GB126 tok/s1214K ctx
dense
Qwen3.5 9B Uncensored HauhauCS Aggressive
C469B17.0 GB126 tok/s1214K ctx
dense
SOLAR 10.7B Instruct v1.0 uncensored
C4610.7B18.3 GB150 tok/s1008K ctx
dense
Qwen3.5 9B
C469B17.0 GB126 tok/s1214K ctx
dense
Starling LM 7B7B16.7 GB98 tok/s8K ctx
dense
Nous Hermes 2 SOLAR 10.7B
C4610.7B18.3 GB150 tok/s1008K ctx
dense
Meta Llama 3.1 8B Instruct
C468B16.3 GB112 tok/s1376K ctx
dense
llava llama 3 8b v1 1
C468B16.3 GB112 tok/s1376K ctx
dense
DeepSeek R1 0528 Qwen3 8B8B16.3 GB112 tok/s1376K ctx
dense
Yi Coder 9B Chat
C469B17.0 GB126 tok/s1214K ctx
dense
Meta Llama 3 8B Instruct
C468B16.3 GB112 tok/s1376K ctx
dense
Gemma 2 2B2B13.3 GB28 tok/s8K ctx
dense
DeepSeek R1 Distill Llama 8B8B16.3 GB112 tok/s1376K ctx
dense
glm 4 9b chat 1m
C469B17.0 GB126 tok/s1214K ctx
dense
Dolphin3.0 Llama3.1 8B
C468B16.3 GB112 tok/s1376K ctx
dense
Llama 3 8B Instruct 32k v0.1
C468B16.3 GB112 tok/s1376K ctx
dense
Meta Llama 3.1 8B Instruct
C468B16.3 GB112 tok/s1376K ctx
dense
Llama 2 7B Chat
C467B15.6 GB98 tok/s1584K ctx
dense
vntl llama3 8b v2
C468B16.3 GB112 tok/s1376K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C468B16.3 GB112 tok/s1376K ctx
dense
Mistral 7B Instruct v0.2
C467B15.6 GB98 tok/s1584K ctx
dense
solar finalised finetuned Model 10.7B i1
C4610.7B18.3 GB150 tok/s1008K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C468B16.3 GB112 tok/s1376K ctx
dense
Mistral 7B Instruct v0.3
C467B15.6 GB98 tok/s1584K ctx
dense
Yi 1.5 9B Chat
C469B17.0 GB126 tok/s1214K ctx
dense
SOLAR 10.7B v1.0
C4610.7B18.3 GB150 tok/s1008K ctx
dense
Qwen3 8B DeepSeek v3.2 Speciale Distill
C468B16.3 GB112 tok/s1376K ctx
dense
Hermes 3 Llama 3.1 8B
C468B16.3 GB112 tok/s1376K ctx
dense
Hermes 2 Pro Llama 3 8B
C458B16.3 GB112 tok/s1376K ctx
dense
Dolphin3.0 Llama3.1 8B
C458B16.3 GB112 tok/s1376K ctx
dense
granite 8b code instruct 4k
C458B16.3 GB112 tok/s1376K ctx
dense
dolphin 2.9.4 llama3.1 8b
C458B16.3 GB112 tok/s1376K ctx
dense
Mistral 7B Instruct v0.3
C457B15.6 GB98 tok/s1584K ctx
dense
Hermes 2 Pro Mistral 7B
C457B15.6 GB98 tok/s1584K ctx
dense
openchat 3.6 8b 20240522 IMat
C458B16.3 GB112 tok/s1376K ctx
dense
HelpingAI2.5 10B i1
C4510B17.8 GB140 tok/s1084K ctx
dense
Helply 10.2b chat i1
C4510.2B17.9 GB143 tok/s1061K ctx
dense
Yi 9B Coder i1
C459B17.0 GB126 tok/s1214K ctx
dense
zephyr 7B beta
C457B15.6 GB98 tok/s1584K ctx
dense
aya expanse 8b
C458B16.3 GB112 tok/s1376K ctx
dense
EXAONE 3.5 7.8B Instruct
C457.8B16.2 GB109 tok/s1413K ctx
dense
Nous Hermes 2 Mistral 7B DPO
C457B15.6 GB98 tok/s1584K ctx
dense
dolphin v2 8b abliterated i1
C458B16.3 GB112 tok/s1376K ctx
dense
EXAONE 3.5 7.8B Instruct
C457.8B16.2 GB109 tok/s1413K ctx
dense
HelpingAI2 9B
C459B17.0 GB126 tok/s1214K ctx
dense
Falcon H1R 7B
C457B15.6 GB98 tok/s1584K ctx
dense
EXAONE 3.5 7.8B Instruct i1
C457.8B16.2 GB109 tok/s1413K ctx
dense
aya expanse 8b orthogonal heretic i1
C458B16.3 GB112 tok/s1376K ctx
dense
falcon mamba 7b instruct Q4 K M
C457B15.6 GB98 tok/s1584K ctx
dense
exaone 3.0 7.8b it
C457.8B16.2 GB109 tok/s1413K ctx
dense
zephyr 7B alpha
C457B15.6 GB98 tok/s1584K ctx
dense
speechless zephyr code functionary 7b
C457B15.6 GB98 tok/s1584K ctx
dense
HelpingAI2 9B i1
C459B17.0 GB126 tok/s1214K ctx
dense
HelpingAI 9B 200k i1
C459B17.0 GB126 tok/s1214K ctx
dense
DeepSeek LLM 7B7B22.1 GB98 tok/s4K ctx
dense
Qwen 2.5 Coder 0.5B0.5B11.0 GB7 tok/s131K ctx
dense
StarCoder2 7B
C457B15.6 GB98 tok/s1584K ctx
dense
HelpingAI 9B i1
C459B17.0 GB126 tok/s1214K ctx
dense
japanese stablelm instruct gamma 7B
C457B15.6 GB98 tok/s1584K ctx
dense
aya expanse 8b orthogonal heretic
C458B16.3 GB112 tok/s1376K ctx
dense
Falcon H1 7B Instruct
C457B15.6 GB98 tok/s1584K ctx
dense
Falcon H1R 7B7B15.6 GB98 tok/s1584K ctx
dense
internlm2 5 1 8b chat i1
C458B16.3 GB112 tok/s1376K ctx
dense
internlm3 8b instruct abliterated i1
C458B16.3 GB112 tok/s1376K ctx
dense
Yi 1.5 6B Chat
C456B14.9 GB84 tok/s1862K ctx
dense
Mamba Codestral 7B v0.1
C457B15.6 GB98 tok/s1584K ctx
dense
stabilityai japanese stablelm base gamma 7b
C457B15.6 GB98 tok/s1584K ctx
dense
baichuan2 7b chat
C457B15.6 GB98 tok/s1584K ctx
dense
DiscoPOP zephyr 7b gemma
C457B15.6 GB98 tok/s1584K ctx
dense
stablelm 2 zephyr 1 6b
C456B14.9 GB84 tok/s1862K ctx
dense
starcoder2 7b
C457B15.6 GB98 tok/s1584K ctx
dense
zephyr 7b beta Mistral 7B Instruct v0.2
C457B15.6 GB98 tok/s1584K ctx
dense
OpenChat 3.5 7B Qwen v2.0 i1
C457B15.6 GB98 tok/s1584K ctx
dense
OpenChat 3.5 7B Starling v2.0 i1
C457B15.6 GB98 tok/s1584K ctx
dense
internlm2 math plus 7b IMat
C457B15.6 GB98 tok/s1584K ctx
dense
CodeNinja 1.0 OpenChat 7B i1
C457B15.6 GB98 tok/s1584K ctx
dense
Neural Chat 7B7B16.7 GB98 tok/s8K ctx
dense
internlm2 5 7b chat i1
C457B15.6 GB98 tok/s1584K ctx
dense
OpenSafetyLab MD Judge v0 2 internlm2 7b
C457B15.6 GB98 tok/s1584K ctx
dense
MD Judge v0 2 internlm2 7b i1
C457B15.6 GB98 tok/s1584K ctx
dense
zephyr 7b gemma sft african ultrachat 100k
C457B15.6 GB98 tok/s1584K ctx
dense
jointpreferences mistral 7b sft helpful
C457B15.6 GB98 tok/s1584K ctx
dense
zephyr 7b dpo full i1
C457B15.6 GB98 tok/s1584K ctx
dense
blossom v3 baichuan2 7b i1
C457B15.6 GB98 tok/s1584K ctx
dense
blossom v1 baichuan 7b i1
C457B15.6 GB98 tok/s1584K ctx
dense
Yi 1.5 6B6B15.1 GB84 tok/s4K ctx
dense
Yi 1.5 6B Chat
C456B14.9 GB84 tok/s1862K ctx
dense
stablelm 2 1 6b chat imatrix
C456B14.9 GB84 tok/s1862K ctx
dense
StarCoder2 7B7B15.3 GB98 tok/s16K ctx
dense
HelpingAI2 6B
C446B14.9 GB84 tok/s1862K ctx
dense
HelpingAI2 6B i1
C446B14.9 GB84 tok/s1862K ctx
dense
Nemotron Mini 4B4B14.9 GB56 tok/s4K ctx
dense
HelpingAI2.5 5B i1
C445B14.1 GB70 tok/s2251K ctx
dense
Qwen3.5 4B4B13.4 GB56 tok/s2835K ctx
dense
gemma 3 4b it
C434B13.4 GB56 tok/s2835K ctx
dense
gemma 3 4b it
C434B13.4 GB56 tok/s2835K ctx
dense
Qwen 3 0.6B0.6B11.7 GB8 tok/s33K ctx
dense
Llama 3.2 1B1B11.6 GB14 tok/s128K ctx
dense
Llama 3.2 3B Instruct
C433B13.0 GB42 tok/s3793K ctx
dense
Qwen2.5 3B Instruct3B12.7 GB42 tok/s3808K ctx
dense
Llama 3.2 3B Instruct
C423B12.7 GB42 tok/s3808K ctx
dense
Ministral 3 3B Instruct 2512
C423B12.7 GB42 tok/s3808K ctx
dense
HELVETE 3B
C423B12.7 GB42 tok/s3808K ctx
dense
Hermes 3 Llama 3.2 3B
C423B12.7 GB42 tok/s3808K ctx
dense
stablelm zephyr 3b
C423B12.7 GB42 tok/s3808K ctx
dense
StarCoder2 3B
C423B12.7 GB42 tok/s3808K ctx
dense
AI21 Jamba Reasoning 3B
C423B12.7 GB42 tok/s3808K ctx
dense
stablelm 3b 4e1t
C423B12.7 GB42 tok/s3808K ctx
dense
ai21labs AI21 Jamba Reasoning 3B
C423B12.7 GB42 tok/s3808K ctx
dense
ai21labs AI21 Jamba2 3B
C423B12.7 GB42 tok/s3808K ctx
dense
HelpingAI 3B hindi i1
C423B12.7 GB42 tok/s3808K ctx
dense
AI21 Jamba2 3B
C423B12.7 GB42 tok/s3808K ctx
dense
HelpingAI 3B hindi
C423B12.7 GB42 tok/s3808K ctx
dense
AI21 Jamba2 3B i1
C423B12.7 GB42 tok/s3808K ctx
dense
gemma 2 2b it
C412B12.4 GB28 tok/s5725K ctx
dense
EXAONE 3.5 2.4B Instruct
C412.4B12.2 GB34 tok/s4781K ctx
dense
gemma 2b2B12.0 GB28 tok/s5754K ctx
dense
Gemmasutra Mini 2B v1
C412B12.0 GB28 tok/s5754K ctx
dense
gemma 2 2b it
C412B12.0 GB28 tok/s5754K ctx
dense
Qwen 3.5 0.6B0.6B11.7 GB8 tok/s131K ctx
dense
StarCoder2 3B3B12.8 GB42 tok/s16K ctx
dense
F
F Tier -- Can't Run (44 models)
Too heavy for this hardware.
Qwen2.5 1.5B Instruct1.5B11.6 GB21 tok/s7699K ctx
dense
DeepSeek R1 Distill Qwen 1.5B1.5B11.6 GB21 tok/s7699K ctx
dense
Yi Coder 1.5B Chat
D401.5B11.6 GB21 tok/s7699K ctx
dense
stablelm 2 zephyr 1.6b
D401.6B11.7 GB22 tok/s7213K ctx
dense
logos16v2 stablelm2 1.6b i1
D401.6B11.7 GB22 tok/s7213K ctx
dense
Yi Coder 1.5B
D391.5B11.6 GB21 tok/s7699K ctx
dense
Falcon H1 1.5B Instruct1.5B11.6 GB21 tok/s7699K ctx
dense
TinyLlama 1.1B Chat v1.0
D391.1B11.3 GB15 tok/s10529K ctx
dense
Qwen 2.5 0.5B0.5B11.0 GB7 tok/s131K ctx
dense
EXAONE 4.0 1.2B
D391.2B11.4 GB17 tok/s9645K ctx
dense
Llama 3.2 1B Instruct Q8 0
D391B11.4 GB14 tok/s11562K ctx
dense
TinyLlama 1.1B Chat v0.3
D391.1B11.3 GB15 tok/s10529K ctx
dense
TinyLlama 1.1B Chat v0.61.1B11.3 GB15 tok/s10529K ctx
dense
Llama 3.2 1B Instruct
D381B11.2 GB14 tok/s11590K ctx
dense
gemma 3 1b it
D381B11.2 GB14 tok/s11590K ctx
dense
TinyLlama 1.1B Chat v1.0 imatrix
D381.1B11.3 GB15 tok/s10529K ctx
dense
Falcon3 1B Instruct abliterated
D381B11.2 GB14 tok/s11590K ctx
dense
SmolVLM 500M Instruct
D360.5B11.0 GB7 tok/s13614K ctx
dense
embeddinggemma 300M
D350.3B10.8 GB4 tok/s13641K ctx
dense
granite embedding 107m multilingual
D330.11B10.7 GB2 tok/s13670K ctx
dense
Falcon H1 Tiny 90M Instruct
D330.09B10.7 GB2 tok/s13671K ctx
dense
Qwen 3 235B A22B235B156.7 GB25 tok/s4K ctx
moe
Qwen3-Coder 480B A35B Instruct480B306.2 GB5 tok/s4K ctx
moe
Qwen 3.5 397B A17B397B255.5 GB7 tok/s4K ctx
moe
Mistral Large 3675B427.6 GB4 tok/s4K ctx
moe
GLM-5744B483.4 GB3 tok/s4K ctx
moe
GLM-5.1754B489.5 GB3 tok/s4K ctx
moe
Kimi K2.51000B627.9 GB3 tok/s4K ctx
moe
Kimi K2.61000B627.9 GB3 tok/s4K ctx
moe
DeepSeek V3 671B671B479.4 GB4 tok/s4K ctx
moe
DeepSeek V3.1 671B671B479.4 GB4 tok/s4K ctx
moe
DeepSeek V4 Pro1600B874.4 GB2 tok/s4K ctx
moe
DeepSeek V4 Flash284B169.8 GB23 tok/s4K ctx
moe
DeepSeek R1 671B671B479.4 GB4 tok/s4K ctx
moe
DeepSeek Coder V2 236B236B213.1 GB14 tok/s4K ctx
moe
DeepSeek V2.5 236B236B213.1 GB14 tok/s4K ctx
moe
Llama 4 Maverick 17B 128E400B257.4 GB7 tok/s4K ctx
moe
Llama 3.1 405B405B265.2 GB2 tok/s4K ctx
dense
DeepSeek V3.2671B420.3 GB4 tok/s4K ctx
moe
MiniMax M2.7
F0230B154.6 GB29 tok/s4K ctx
moe
K EXAONE 236B A23B
F0236B182.1 GB7 tok/s4K ctx
dense
Baichuan M3 235B
F0235B181.4 GB7 tok/s4K ctx
dense
Baichuan M3 235B i1
F0235B181.4 GB7 tok/s4K ctx
dense
Qwen3.5 397B A17B397B299.2 GB2 tok/s4K ctx
dense
Compare with...
See How Other Hardware Stacks Up
Check tier lists for similar GPUs and Apple Silicon configs