AI Compatibility Tier List
Mac Studio M3 Ultra 256GB
374 AI models ranked by how well they run on your 256 GB unified memory hardware. Best match: Qwen 3.5 122B A10B (score: 93).
374
Models tested
340
Can run
256 GB
unified memory
35
Best tok/s
Tier Distribution
S 28
A 56
B 60
C 196
F 34
S: 28A: 56B: 60C: 196F: 34
S
S Tier -- Perfect Fit (28 models)
These models run flawlessly with headroom to spare.
Qwen 3.5 122B A10B122B105.4 GB35 tok/s131K ctx
moe
Mistral Small 4 119B119B106.5 GB38 tok/s248K ctx
moe
DeepSeek V4 Flash284B187.9 GB18 tok/s4K ctx
moe
Qwen3-Coder 30B A3B Instruct30.5B48.6 GB84 tok/s256K ctx
moe
Devstral 2 123B Instruct123B108.9 GB8 tok/s241K ctx
dense
Qwen3-VL 30B A3B Instruct30B48.3 GB87 tok/s256K ctx
moe
Qwen 3.6 35B A3B35B54.0 GB71 tok/s262K ctx
moe
Qwen 3.5 35B A3B35B51.4 GB77 tok/s131K ctx
moe
Qwen3-Coder-Next80B78.8 GB38 tok/s256K ctx
moe
Qwen 3 30B A3B30.5B48.6 GB84 tok/s131K ctx
moe
Qwen 3.5 9B9B36.2 GB109 tok/s131K ctx
dense
Leanstral 119B A6B119B109.9 GB35 tok/s151K ctx
moe
Qwen 3.5 27B27B48.2 GB37 tok/s131K ctx
dense
GPT-OSS 120B117B104.8 GB9 tok/s131K ctx
dense
Command A 111B111B100.2 GB9 tok/s262K ctx
dense
Pixtral Large 124B124B109.6 GB8 tok/s131K ctx
dense
Qwen 3 14B14B39.5 GB70 tok/s131K ctx
dense
Magistral Small 250724B45.6 GB41 tok/s131K ctx
dense
Qwen 3 8B8B35.6 GB112 tok/s131K ctx
dense
Devstral Small 2 24B Instruct24B45.6 GB41 tok/s256K ctx
dense
Qwen 3.6 27B27B46.0 GB28 tok/s262K ctx
dense
Qwen 3 235B A22B235B174.8 GB11 tok/s69K ctx
moe
GPT-OSS 20B21B43.8 GB107 tok/s128K ctx
moe
Qwen 3 32B32B52.0 GB31 tok/s131K ctx
dense
MiniMax M2.7
S85230B172.6 GB20 tok/s65K ctx
moe
Phi-4-reasoning-plus 14B14.7B40.6 GB67 tok/s33K ctx
dense
Qwen 2.5 VL 72B72B77.4 GB14 tok/s33K ctx
dense
Nemotron Cascade 2 30B A3B30B49.8 GB86 tok/s262K ctx
moe
A
A Tier -- Great Fit (56 models)
Excellent performance with comfortable memory margins.
Nemotron 3 Nano 30B30B49.3 GB33 tok/s131K ctx
dense
Qwen 3.5 4B4B33.2 GB56 tok/s131K ctx
dense
Devstral Small 1.124B45.6 GB41 tok/s131K ctx
dense
Gemma 4 26B A4B25.2B47.6 GB90 tok/s256K ctx
moe
Gemma 4 31B30.7B61.9 GB24 tok/s150K ctx
dense
Nemotron Nano 8B8B35.4 GB112 tok/s131K ctx
dense
Phi-4 Mini Reasoning 4B3.8B32.3 GB53 tok/s131K ctx
dense
Ministral 3 14B14B39.5 GB70 tok/s262K ctx
multimodal
EXAONE 4.0 32B32B52.0 GB31 tok/s131K ctx
dense
Llama 3.3 70B70B76.1 GB14 tok/s128K ctx
dense
Codestral 2 25.0822B44.4 GB42 tok/s256K ctx
dense
Mistral Small 3.2 24B24B45.6 GB41 tok/s131K ctx
vision
InternVL2 8B8B35.4 GB112 tok/s8K ctx
dense
CogVLM2 19B19B42.6 GB52 tok/s8K ctx
dense
Qwen 2.5 32B32B52.0 GB31 tok/s131K ctx
dense
LFM2 24B24B45.6 GB41 tok/s131K ctx
dense
Phi-4 14B14B40.1 GB70 tok/s16K ctx
dense
Gemma 3 27B27B56.2 GB27 tok/s131K ctx
dense
Granite 4.1 30B30B50.8 GB33 tok/s131K ctx
dense
OLMo 2 32B32B52.0 GB31 tok/s4K ctx
dense
Ministral 3 8B8B35.6 GB112 tok/s262K ctx
multimodal
Mistral Small 24B24B45.6 GB41 tok/s33K ctx
dense
MiniCPM-V 2.6 8B8B35.4 GB112 tok/s2K ctx
dense
Qwen 2.5 14B14B40.0 GB70 tok/s131K ctx
dense
Llama 4 Scout 17B 16E109B98.0 GB21 tok/s488K ctx
moe
Kimi Linear 48B A3B48B58.8 GB19 tok/s1049K ctx
linear
Llama 3.1 70B70B76.1 GB14 tok/s128K ctx
dense
Qwen 2.5 72B72B77.4 GB14 tok/s131K ctx
dense
Mistral Small 3.1 24B24B45.6 GB41 tok/s131K ctx
dense
Qwen 3 4B4B33.2 GB56 tok/s33K ctx
dense
Nemotron Nano 9B v29B36.5 GB109 tok/s131K ctx
dense
SQLCoder 7B7B34.8 GB98 tok/s8K ctx
dense
Qwen 2.5 VL 7B7B33.7 GB98 tok/s33K ctx
dense
DeepSeek Coder V2 16B16B41.6 GB136 tok/s131K ctx
moe
Gemma 3 12B12B40.8 GB61 tok/s131K ctx
dense
Jina Embeddings v30.57B31.7 GB8 tok/s8K ctx
dense
Granite Code 20B20B43.9 GB49 tok/s8K ctx
dense
Magistral 7B7B34.8 GB98 tok/s8K ctx
dense
CodeGeeX 4 9B9B34.6 GB111 tok/s131K ctx
dense
BGE M30.57B30.8 GB8 tok/s8K ctx
dense
Gemma 4 E4B8B34.7 GB93 tok/s128K ctx
dense
OLMo 2 13B13B38.9 GB76 tok/s33K ctx
dense
Qwen 2.5 Coder 32B32B52.0 GB31 tok/s131K ctx
dense
CodeLlama 13B Instruct13B48.7 GB70 tok/s16K ctx
dense
Qwen 2.5 7B7B33.7 GB98 tok/s131K ctx
dense
StarCoder 15B15B54.0 GB53 tok/s8K ctx
dense
Granite Code 34B34B53.0 GB29 tok/s8K ctx
dense
Granite Code 8B8B35.4 GB112 tok/s8K ctx
dense
DeepSeek R1 Distill 70B70B76.1 GB14 tok/s131K ctx
dense
DevStral 7B7B34.8 GB98 tok/s8K ctx
dense
Codestral Mamba 7B7B33.3 GB98 tok/s262K ctx
state-space
Command R 35B35B52.3 GB28 tok/s131K ctx
dense
mxbai Embed Large0.34B30.7 GB5 tok/s1K ctx
dense
Granite 4.1 8B8B35.9 GB112 tok/s131K ctx
dense
DeepSeek R1 Distill 14B14B40.0 GB70 tok/s33K ctx
dense
LLaVA 1.6 13B13B48.7 GB70 tok/s4K ctx
dense
B
B Tier -- Good Fit (60 models)
Solid performance, may need quantization or reduced context.
CodeLlama 7B Instruct7B40.6 GB98 tok/s16K ctx
dense
StarCoder 7B7B40.1 GB98 tok/s8K ctx
dense
DeepSeek R1 Distill 32B32B52.0 GB31 tok/s33K ctx
dense
Pixtral 12B12B38.3 GB82 tok/s131K ctx
dense
Snowflake Arctic Embed L0.34B30.7 GB5 tok/s1K ctx
dense
Qwen 2.5 Coder 3B3B32.6 GB42 tok/s131K ctx
dense
Nous Hermes 1.09B46.2 GB101 tok/s16K ctx
dense
Gemma 4 E2B5.1B32.2 GB71 tok/s128K ctx
dense
BGE Large EN v1.50.34B30.7 GB5 tok/s1K ctx
dense
WizardLM 13B13B48.7 GB70 tok/s8K ctx
dense
Vicuna 13B13B48.7 GB70 tok/s4K ctx
dense
GLM-4 9B9B34.6 GB111 tok/s128K ctx
dense
InternLM 7B7B40.6 GB98 tok/s8K ctx
dense
InternLM Chat 7B7B40.6 GB98 tok/s8K ctx
dense
MPT-30B-Instruct30B73.6 GB26 tok/s8K ctx
dense
Llama 3.1 8B8B35.4 GB112 tok/s128K ctx
dense
Nous Dolphin 13B13B50.1 GB61 tok/s16K ctx
dense
Solar 7B7B35.7 GB98 tok/s8K ctx
dense
WizardMath 7B7B34.8 GB98 tok/s4K ctx
dense
OLMo 2 7B7B34.8 GB98 tok/s4K ctx
dense
Nemotron 70B70B76.1 GB14 tok/s131K ctx
dense
Nomic Embed Text v1.50.14B29.4 GB2 tok/s8K ctx
dense
Qwen 2.5 Coder 7B7B33.7 GB98 tok/s131K ctx
dense
Ministral 3 3B3B31.1 GB42 tok/s262K ctx
multimodal
Gemma 3 4B4B33.1 GB56 tok/s128K ctx
dense
Falcon 40B Instruct40B59.2 GB22 tok/s8K ctx
dense
Phi 4 Mini 4B4B32.5 GB56 tok/s128K ctx
dense
LLaVA 1.5 7B7B40.6 GB98 tok/s4K ctx
dense
Gemma 2 27B27B56.2 GB27 tok/s8K ctx
dense
Qwen 3.5 2B2B31.5 GB28 tok/s131K ctx
dense
Command R+ 104B104B95.4 GB10 tok/s131K ctx
dense
Mixtral 8x22B141B118.0 GB14 tok/s66K ctx
moe
DeepSeek R1 Distill 7B7B33.7 GB98 tok/s33K ctx
dense
MPT-7B-Instruct7B40.6 GB98 tok/s8K ctx
dense
Baichuan 7B7B40.6 GB98 tok/s8K ctx
dense
DeepSeek R1 Distill 8B8B35.4 GB112 tok/s33K ctx
dense
Baichuan 13B13B50.1 GB61 tok/s8K ctx
dense
Cerebras-GPT 13B13B47.7 GB61 tok/s131K ctx
dense
Samantha 7B7B34.8 GB98 tok/s4K ctx
dense
Qwen 2.5 3B3B32.6 GB42 tok/s131K ctx
dense
Falcon 7B Instruct7B32.9 GB98 tok/s8K ctx
dense
Mixtral 8x7B47B59.2 GB40 tok/s33K ctx
moe
Phi 3 Mini 3.8B3.8B36.7 GB53 tok/s128K ctx
dense
Qwen 3 1.7B1.7B31.3 GB24 tok/s33K ctx
dense
All MiniLM L6 v20.02B28.9 GB2 tok/s0K ctx
dense
Llama 3.2 11B Vision11B37.2 GB89 tok/s16K ctx
vision
Gemma 2 9B9B39.2 GB81 tok/s8K ctx
dense
Phi 3.5 Mini 4B4B36.8 GB56 tok/s128K ctx
dense
Granite 4.1 3B3B31.6 GB42 tok/s131K ctx
dense
Qwen 2.5 Coder 14B14B40.0 GB70 tok/s131K ctx
dense
Qwen 2.5 Coder 1.5B1.5B29.9 GB21 tok/s33K ctx
dense
Granite Code 3B3B32.8 GB42 tok/s8K ctx
dense
Mistral Nemo 12B12B38.3 GB82 tok/s128K ctx
dense
Yi Coder 9B9B35.5 GB110 tok/s131K ctx
dense
Qwen 2.5 Math 72B72B77.4 GB14 tok/s4K ctx
dense
Mistral 7B Instruct v0.37B34.8 GB98 tok/s8K ctx
dense
Yi 1.5 34B34B53.0 GB29 tok/s4K ctx
dense
Phi 3 Medium 14B14B40.1 GB70 tok/s128K ctx
dense
Llama 3.2 3B3B32.1 GB42 tok/s128K ctx
dense
Ministral 8B8B35.6 GB112 tok/s131K ctx
dense
C
C Tier -- Tight Fit (196 models)
Runs with offloading or significant compromises.
Codestral 22B22B44.4 GB45 tok/s33K ctx
dense
DeepSeek LLM 67B67B75.2 GB15 tok/s4K ctx
dense
InternLM 20B20B61.3 GB39 tok/s8K ctx
dense
DeepSeek R1 1.5B1.5B29.9 GB21 tok/s33K ctx
dense
TinyLlama 1.1B1.1B29.6 GB15 tok/s4K ctx
dense
SmolLM3 3B3B32.3 GB42 tok/s128K ctx
dense
Granite 3.1 8B8B35.4 GB112 tok/s128K ctx
state-space
Yi 1.5 9B9B35.5 GB110 tok/s4K ctx
dense
Qwen 2.5 Math 7B7B33.7 GB98 tok/s4K ctx
dense
Aya Expanse 32B32B50.5 GB31 tok/s8K ctx
dense
Qwen 2.5 1.5B1.5B29.9 GB21 tok/s131K ctx
dense
OpenChat 7B7B34.8 GB98 tok/s8K ctx
dense
Aya Expanse 8B8B35.4 GB112 tok/s8K ctx
dense
Gemma 3 1B1B29.6 GB14 tok/s33K ctx
dense
Vicuna 7B7B40.6 GB98 tok/s4K ctx
dense
OpenHermes 2.5 7B7B34.8 GB98 tok/s8K ctx
dense
Qwen3.5 122B A10B122B102.6 GB9 tok/s107K ctx
dense
Dolphin 2.9 8B8B35.4 GB112 tok/s33K ctx
dense
Zephyr 7B Beta7B34.8 GB98 tok/s33K ctx
dense
StarCoder2 15B15B40.6 GB57 tok/s16K ctx
dense
Gemma 2 2B2B31.4 GB28 tok/s8K ctx
dense
Qwen3.5 9B9B35.1 GB101 tok/s2280K ctx
dense
Qwen3.5 9B Uncensored HauhauCS Aggressive
C469B35.1 GB101 tok/s2280K ctx
dense
Starling LM 7B7B34.8 GB98 tok/s8K ctx
dense
Meta Llama 3.1 8B Instruct
C468B34.4 GB112 tok/s2575K ctx
dense
Qwen3.5 9B
C469B35.1 GB101 tok/s2280K ctx
dense
llava llama 3 8b v1 1
C468B34.4 GB112 tok/s2575K ctx
dense
DeepSeek R1 0528 Qwen3 8B8B34.4 GB112 tok/s2575K ctx
dense
Solar Open 100B
C46100B101.3 GB9 tok/s129K ctx
dense
Solar Open 100B i1
C46100B101.3 GB9 tok/s129K ctx
dense
Llama 2 7B Chat
C467B33.6 GB98 tok/s2955K ctx
dense
Meta Llama 3 8B Instruct
C468B34.4 GB112 tok/s2575K ctx
dense
Qwen 2.5 Coder 0.5B0.5B29.0 GB7 tok/s131K ctx
dense
DeepSeek R1 Distill Llama 8B8B34.4 GB112 tok/s2575K ctx
dense
Dolphin3.0 Llama3.1 8B
C468B34.4 GB112 tok/s2575K ctx
dense
Llama 3 8B Instruct 32k v0.1
C468B34.4 GB112 tok/s2575K ctx
dense
Mistral 7B Instruct v0.2
C467B33.6 GB98 tok/s2955K ctx
dense
Meta Llama 3.1 8B Instruct
C468B34.4 GB112 tok/s2575K ctx
dense
vntl llama3 8b v2
C468B34.4 GB112 tok/s2575K ctx
dense
StableLM 2 12B12B49.4 GB60 tok/s4K ctx
dense
Yi Coder 9B Chat
C469B35.1 GB101 tok/s2280K ctx
dense
Mistral 7B Instruct v0.3
C467B33.6 GB98 tok/s2955K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C468B34.4 GB112 tok/s2575K ctx
dense
glm 4 9b chat 1m
C469B35.1 GB101 tok/s2280K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C468B34.4 GB112 tok/s2575K ctx
dense
Qwen3 8B DeepSeek v3.2 Speciale Distill
C468B34.4 GB112 tok/s2575K ctx
dense
Hermes 3 Llama 3.1 8B
C468B34.4 GB112 tok/s2575K ctx
dense
Yi 34B Chat34B53.0 GB29 tok/s200K ctx
dense
Yi 1.5 9B Chat
C469B35.1 GB101 tok/s2280K ctx
dense
SOLAR 10.7B Instruct v1.0 uncensored
C4610.7B36.3 GB85 tok/s1904K ctx
dense
Hermes 2 Pro Llama 3 8B
C468B34.4 GB112 tok/s2575K ctx
dense
Dolphin3.0 Llama3.1 8B
C468B34.4 GB112 tok/s2575K ctx
dense
Nous Hermes 2 SOLAR 10.7B
C4610.7B36.3 GB85 tok/s1904K ctx
dense
granite 8b code instruct 4k
C468B34.4 GB112 tok/s2575K ctx
dense
Mistral 7B Instruct v0.3
C467B33.6 GB98 tok/s2955K ctx
dense
Hermes 2 Pro Mistral 7B
C467B33.6 GB98 tok/s2955K ctx
dense
dolphin 2.9.4 llama3.1 8b
C468B34.4 GB112 tok/s2575K ctx
dense
gemma 3 12b it
C4612B37.3 GB76 tok/s1689K ctx
dense
zephyr 7B beta
C467B33.6 GB98 tok/s2955K ctx
dense
Nous Hermes 2 Mistral 7B DPO
C467B33.6 GB98 tok/s2955K ctx
dense
openchat 3.6 8b 20240522 IMat
C468B34.4 GB112 tok/s2575K ctx
dense
Falcon H1R 7B
C467B33.6 GB98 tok/s2955K ctx
dense
aya expanse 8b
C468B34.4 GB112 tok/s2575K ctx
dense
EXAONE 3.5 7.8B Instruct
C467.8B34.2 GB109 tok/s2643K ctx
dense
falcon mamba 7b instruct Q4 K M
C467B33.6 GB98 tok/s2955K ctx
dense
dolphin v2 8b abliterated i1
C468B34.4 GB112 tok/s2575K ctx
dense
zephyr 7B alpha
C467B33.6 GB98 tok/s2955K ctx
dense
EXAONE 3.5 7.8B Instruct
C467.8B34.2 GB109 tok/s2643K ctx
dense
speechless zephyr code functionary 7b
C467B33.6 GB98 tok/s2955K ctx
dense
Yi 9B Coder i1
C469B35.1 GB101 tok/s2280K ctx
dense
EXAONE 3.5 7.8B Instruct i1
C467.8B34.2 GB109 tok/s2643K ctx
dense
StarCoder2 7B
C467B33.6 GB98 tok/s2955K ctx
dense
aya expanse 8b orthogonal heretic i1
C468B34.4 GB112 tok/s2575K ctx
dense
exaone 3.0 7.8b it
C467.8B34.2 GB109 tok/s2643K ctx
dense
japanese stablelm instruct gamma 7B
C467B33.6 GB98 tok/s2955K ctx
dense
solar finalised finetuned Model 10.7B i1
C4610.7B36.3 GB85 tok/s1904K ctx
dense
HelpingAI2 9B
C469B35.1 GB101 tok/s2280K ctx
dense
Falcon H1 7B Instruct
C467B33.6 GB98 tok/s2955K ctx
dense
HelpingAI2.5 10B i1
C4610B35.8 GB91 tok/s2044K ctx
dense
Falcon H1R 7B7B33.6 GB98 tok/s2955K ctx
dense
Yi 1.5 6B Chat
C466B32.9 GB84 tok/s3461K ctx
dense
SOLAR 10.7B v1.0
C4510.7B36.3 GB85 tok/s1904K ctx
dense
aya expanse 8b orthogonal heretic
C458B34.4 GB112 tok/s2575K ctx
dense
HelpingAI2 9B i1
C459B35.1 GB101 tok/s2280K ctx
dense
Helply 10.2b chat i1
C4510.2B36.0 GB90 tok/s2002K ctx
dense
HelpingAI 9B 200k i1
C459B35.1 GB101 tok/s2280K ctx
dense
stablelm 2 zephyr 1 6b
C456B32.9 GB84 tok/s3461K ctx
dense
HelpingAI 9B i1
C459B35.1 GB101 tok/s2280K ctx
dense
Mamba Codestral 7B v0.1
C457B33.6 GB98 tok/s2955K ctx
dense
DeepSeek R1 Distill Qwen 14B14B38.7 GB65 tok/s1436K ctx
dense
stabilityai japanese stablelm base gamma 7b
C457B33.6 GB98 tok/s2955K ctx
dense
baichuan2 7b chat
C457B33.6 GB98 tok/s2955K ctx
dense
internlm2 5 1 8b chat i1
C458B34.4 GB112 tok/s2575K ctx
dense
internlm3 8b instruct abliterated i1
C458B34.4 GB112 tok/s2575K ctx
dense
DiscoPOP zephyr 7b gemma
C457B33.6 GB98 tok/s2955K ctx
dense
starcoder2 7b
C457B33.6 GB98 tok/s2955K ctx
dense
zephyr 7b beta Mistral 7B Instruct v0.2
C457B33.6 GB98 tok/s2955K ctx
dense
OpenChat 3.5 7B Qwen v2.0 i1
C457B33.6 GB98 tok/s2955K ctx
dense
OpenChat 3.5 7B Starling v2.0 i1
C457B33.6 GB98 tok/s2955K ctx
dense
internlm2 math plus 7b IMat
C457B33.6 GB98 tok/s2955K ctx
dense
CodeNinja 1.0 OpenChat 7B i1
C457B33.6 GB98 tok/s2955K ctx
dense
internlm2 5 7b chat i1
C457B33.6 GB98 tok/s2955K ctx
dense
OpenSafetyLab MD Judge v0 2 internlm2 7b
C457B33.6 GB98 tok/s2955K ctx
dense
Yi 1.5 6B6B33.2 GB84 tok/s4K ctx
dense
MD Judge v0 2 internlm2 7b i1
C457B33.6 GB98 tok/s2955K ctx
dense
Yi 1.5 6B Chat
C456B32.9 GB84 tok/s3461K ctx
dense
stablelm 2 1 6b chat imatrix
C456B32.9 GB84 tok/s3461K ctx
dense
zephyr 7b gemma sft african ultrachat 100k
C457B33.6 GB98 tok/s2955K ctx
dense
jointpreferences mistral 7b sft helpful
C457B33.6 GB98 tok/s2955K ctx
dense
zephyr 7b dpo full i1
C457B33.6 GB98 tok/s2955K ctx
dense
blossom v3 baichuan2 7b i1
C457B33.6 GB98 tok/s2955K ctx
dense
blossom v1 baichuan 7b i1
C457B33.6 GB98 tok/s2955K ctx
dense
Neural Chat 7B7B34.8 GB98 tok/s8K ctx
dense
NousResearch Hermes 4 14B
C4514B38.7 GB65 tok/s1436K ctx
dense
Phi 4 reasoning vision 15B
C4515B39.5 GB61 tok/s1335K ctx
dense
StarCoder2 15B
C4515B39.5 GB61 tok/s1335K ctx
dense
Llama 3.3 70B Instruct
C4570B79.5 GB13 tok/s221K ctx
dense
HelpingAI2 6B
C456B32.9 GB84 tok/s3461K ctx
dense
StarCoder2 7B7B33.3 GB98 tok/s16K ctx
dense
GGUF SOLARized GraniStral 14B 2102 YeAM HCT 32QKV
C4514B38.7 GB65 tok/s1436K ctx
dense
HelpingAI2 6B i1
C456B32.9 GB84 tok/s3461K ctx
dense
Nemotron Mini 4B4B32.9 GB56 tok/s4K ctx
dense
internlm JanusCoder 14B
C4514B38.7 GB65 tok/s1436K ctx
dense
DeepSeek LLM 7B7B40.1 GB98 tok/s4K ctx
dense
GGUF SOLARized GraniStral 14B 1902 YeAM HCT
C4514B38.7 GB65 tok/s1436K ctx
dense
starcoder2 15b instruct v0.1
C4515B39.5 GB61 tok/s1335K ctx
dense
starcoder2 15b i1
C4515B39.5 GB61 tok/s1335K ctx
dense
starcoder2 15b instruct v0.1
C4515B39.5 GB61 tok/s1335K ctx
dense
HelpingAI 15B i1
C4515B39.5 GB61 tok/s1335K ctx
dense
Qwen3.5 27B27B48.2 GB34 tok/s704K ctx
dense
Solar Open 69B REAP i1
C4569B78.7 GB13 tok/s225K ctx
dense
Mistral Small 3.2 24B Instruct 250624B46.0 GB38 tok/s803K ctx
dense
gemma 3 27b it27B48.2 GB34 tok/s704K ctx
dense
Qwen3.5 35B A3B35B54.0 GB26 tok/s524K ctx
dense
stabilityai japanese stablelm instruct beta 70b
C4470B79.5 GB13 tok/s221K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
C4424B46.0 GB38 tok/s803K ctx
dense
Mistral Small 24B Instruct 2501
C4424B46.0 GB38 tok/s803K ctx
dense
Codestral 22B v0.1
C4422B44.5 GB42 tok/s883K ctx
dense
Codestral 22B v0.1
C4422B44.5 GB42 tok/s883K ctx
dense
BaichuanMed OCR 72B i1
C4472B80.9 GB13 tok/s212K ctx
dense
mistral small 3.1 24b instruct 2503 hf
C4424B46.0 GB38 tok/s803K ctx
dense
cognitivecomputations Dolphin3.0 R1 Mistral 24B
C4424B46.0 GB38 tok/s803K ctx
dense
gemma 3 27b it
C4427B48.2 GB34 tok/s704K ctx
dense
Qwen3.5 35B A3B
C4435B54.0 GB26 tok/s524K ctx
dense
Dolphin Mistral GLM 4.7 Flash 24B Venice Edition Thinking Uncensored i1
C4424B46.0 GB38 tok/s803K ctx
dense
HelpingAI2.5 5B i1
C445B32.2 GB70 tok/s4170K ctx
dense
Qwen3.5 4B4B31.5 GB56 tok/s5234K ctx
dense
Codestral 22B v0.1
C4422B44.5 GB42 tok/s883K ctx
dense
internlm2 5 20b chat
C4420B43.1 GB46 tok/s980K ctx
dense
internlm2 limarp chat 20b
C4420B43.1 GB46 tok/s980K ctx
dense
Codestral RAG 19B Pruned i1
C4419B42.4 GB48 tok/s1036K ctx
dense
gemma 3 4b it
C444B31.5 GB56 tok/s5234K ctx
dense
internlm2 math plus 20b i1
C4420B43.1 GB46 tok/s980K ctx
dense
Codestral 22B v0.1 IMat
C4422B44.5 GB42 tok/s883K ctx
dense
Codestral 21B Pruned i1
C4421B43.8 GB44 tok/s929K ctx
dense
gemma 3 4b it
C444B31.5 GB56 tok/s5234K ctx
dense
Qwen3 48B A4B Savant Commander Distill 12X Closed Open Heretic Uncensored
C4448B63.5 GB19 tok/s360K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
C4424B46.0 GB38 tok/s803K ctx
dense
Qwen 3 0.6B0.6B29.8 GB8 tok/s33K ctx
dense
Codestral 22B v0.1 i1
C4422B44.5 GB42 tok/s883K ctx
dense
Hermes 4.3 36B
C4436B54.7 GB25 tok/s507K ctx
dense
Llama 3.2 1B1B29.6 GB14 tok/s128K ctx
dense
EXAONE 4.0 32B
C4432B51.8 GB29 tok/s581K ctx
dense
aya expanse 32b heretic MPOA i1
C4432B51.8 GB29 tok/s581K ctx
dense
Baichuan M2 32B Q4 K M
C4432B51.8 GB29 tok/s581K ctx
dense
baichuan inc Baichuan M2 32B
C4432B51.8 GB29 tok/s581K ctx
dense
Llama 3.2 3B Instruct
C433B31.1 GB42 tok/s6991K ctx
dense
Qwen2.5 3B Instruct3B30.7 GB42 tok/s7006K ctx
dense
Llama 3.2 3B Instruct
C433B30.7 GB42 tok/s7006K ctx
dense
Ministral 3 3B Instruct 2512
C433B30.7 GB42 tok/s7006K ctx
dense
HELVETE 3B
C433B30.7 GB42 tok/s7006K ctx
dense
Hermes 3 Llama 3.2 3B
C433B30.7 GB42 tok/s7006K ctx
dense
stablelm zephyr 3b
C433B30.7 GB42 tok/s7006K ctx
dense
StarCoder2 3B
C433B30.7 GB42 tok/s7006K ctx
dense
AI21 Jamba Reasoning 3B
C433B30.7 GB42 tok/s7006K ctx
dense
stablelm 3b 4e1t
C433B30.7 GB42 tok/s7006K ctx
dense
ai21labs AI21 Jamba Reasoning 3B
C433B30.7 GB42 tok/s7006K ctx
dense
ai21labs AI21 Jamba2 3B
C423B30.7 GB42 tok/s7006K ctx
dense
HelpingAI 3B hindi i1
C423B30.7 GB42 tok/s7006K ctx
dense
AI21 Jamba2 3B
C423B30.7 GB42 tok/s7006K ctx
dense
HelpingAI 3B hindi
C423B30.7 GB42 tok/s7006K ctx
dense
AI21 Jamba2 3B i1
C423B30.7 GB42 tok/s7006K ctx
dense
EXAONE 3.5 2.4B Instruct
C422.4B30.3 GB34 tok/s8778K ctx
dense
gemma 2 2b it
C422B30.4 GB28 tok/s10522K ctx
dense
gemma 2b2B30.0 GB28 tok/s10551K ctx
dense
Gemmasutra Mini 2B v1
C422B30.0 GB28 tok/s10551K ctx
dense
Qwen 3.5 0.6B0.6B29.8 GB8 tok/s131K ctx
dense
gemma 2 2b it
C422B30.0 GB28 tok/s10551K ctx
dense
StarCoder2 3B3B30.8 GB42 tok/s16K ctx
dense
Qwen2.5 1.5B Instruct1.5B29.6 GB21 tok/s14095K ctx
dense
DeepSeek R1 Distill Qwen 1.5B1.5B29.6 GB21 tok/s14095K ctx
dense
Yi Coder 1.5B Chat
C411.5B29.6 GB21 tok/s14095K ctx
dense
stablelm 2 zephyr 1.6b
C401.6B29.7 GB22 tok/s13209K ctx
dense
logos16v2 stablelm2 1.6b i1
C401.6B29.7 GB22 tok/s13209K ctx
dense
Yi Coder 1.5B
C401.5B29.6 GB21 tok/s14095K ctx
dense
Falcon H1 1.5B Instruct1.5B29.6 GB21 tok/s14095K ctx
dense
F
F Tier -- Can't Run (34 models)
Too heavy for this hardware.
TinyLlama 1.1B Chat v1.0
D401.1B29.3 GB15 tok/s19251K ctx
dense
Qwen 2.5 0.5B0.5B29.0 GB7 tok/s131K ctx
dense
EXAONE 4.0 1.2B
D401.2B29.4 GB17 tok/s17640K ctx
dense
Llama 3.2 1B Instruct Q8 0
D401B29.5 GB14 tok/s21156K ctx
dense
TinyLlama 1.1B Chat v0.3
D401.1B29.3 GB15 tok/s19251K ctx
dense
TinyLlama 1.1B Chat v0.61.1B29.3 GB15 tok/s19251K ctx
dense
Llama 3.2 1B Instruct
D391B29.3 GB14 tok/s21185K ctx
dense
gemma 3 1b it
D391B29.3 GB14 tok/s21185K ctx
dense
TinyLlama 1.1B Chat v1.0 imatrix
D391.1B29.3 GB15 tok/s19251K ctx
dense
Falcon3 1B Instruct abliterated
D391B29.3 GB14 tok/s21185K ctx
dense
SmolVLM 500M Instruct
D370.5B29.1 GB7 tok/s24858K ctx
dense
embeddinggemma 300M
D360.3B28.9 GB4 tok/s24884K ctx
dense
granite embedding 107m multilingual
D340.11B28.7 GB2 tok/s24913K ctx
dense
Falcon H1 Tiny 90M Instruct
D340.09B28.7 GB2 tok/s24915K ctx
dense
K EXAONE 236B A23B
D34236B200.2 GB3 tok/s7K ctx
dense
Baichuan M3 235B
D33235B199.4 GB3 tok/s7K ctx
dense
Baichuan M3 235B i1
D33235B199.4 GB3 tok/s7K ctx
dense
Qwen3-Coder 480B A35B Instruct480B324.2 GB4 tok/s4K ctx
moe
Qwen 3.5 397B A17B397B273.6 GB7 tok/s4K ctx
moe
Mistral Large 3675B445.7 GB3 tok/s4K ctx
moe
GLM-5744B501.4 GB3 tok/s4K ctx
moe
GLM-5.1754B507.5 GB3 tok/s4K ctx
moe
Kimi K2.51000B646.0 GB2 tok/s4K ctx
moe
Kimi K2.61000B646.0 GB2 tok/s4K ctx
moe
DeepSeek V3 671B671B497.4 GB3 tok/s4K ctx
moe
DeepSeek V3.1 671B671B497.4 GB3 tok/s4K ctx
moe
DeepSeek V4 Pro1600B892.4 GB2 tok/s4K ctx
moe
DeepSeek R1 671B671B497.4 GB3 tok/s4K ctx
moe
DeepSeek Coder V2 236B236B231.1 GB12 tok/s4K ctx
moe
DeepSeek V2.5 236B236B231.1 GB12 tok/s4K ctx
moe
Llama 4 Maverick 17B 128E400B275.5 GB7 tok/s4K ctx
moe
Llama 3.1 405B405B283.3 GB2 tok/s4K ctx
dense
DeepSeek V3.2671B438.3 GB3 tok/s4K ctx
moe
Qwen3.5 397B A17B397B317.2 GB2 tok/s4K ctx
dense
Compare with...
See How Other Hardware Stacks Up
Check tier lists for similar GPUs and Apple Silicon configs