AI Compatibility Tier List
MacBook Air M4 24GB
374 AI models ranked by how well they run on your 24 GB unified memory hardware. Best match: Qwen 3.5 9B (score: 92).
374
Models tested
257
Can run
24 GB
unified memory
16
Best tok/s
Tier Distribution
S 6
A 40
B 42
C 169
F 117
S: 6A: 40B: 42C: 169F: 117
S
S Tier -- Perfect Fit (6 models)
These models run flawlessly with headroom to spare.
A
A Tier -- Great Fit (40 models)
Excellent performance with comfortable memory margins.
Nemotron Nano 8B8B10.3 GB19 tok/s73K ctx
dense
Ministral 3 14B14B14.5 GB10 tok/s34K ctx
multimodal
InternVL2 8B8B10.3 GB18 tok/s8K ctx
dense
Qwen 3 4B4B8.1 GB35 tok/s33K ctx
dense
Ministral 3 8B8B10.6 GB18 tok/s65K ctx
multimodal
Nemotron Nano 9B v29B11.4 GB17 tok/s54K ctx
dense
CogVLM2 19B19B17.5 GB10 tok/s8K ctx
dense
MiniCPM-V 2.6 8B8B10.3 GB18 tok/s2K ctx
dense
Jina Embeddings v30.57B6.6 GB8 tok/s8K ctx
dense
SQLCoder 7B7B9.7 GB20 tok/s8K ctx
dense
Phi-4 14B14B15.1 GB10 tok/s16K ctx
dense
DeepSeek Coder V2 16B16B16.5 GB21 tok/s20K ctx
moe
Qwen 2.5 14B14B15.0 GB10 tok/s29K ctx
dense
Qwen 2.5 VL 7B7B8.6 GB20 tok/s33K ctx
dense
BGE M30.57B5.8 GB8 tok/s8K ctx
dense
OLMo 2 13B13B13.9 GB10 tok/s33K ctx
dense
Magistral 7B7B9.7 GB20 tok/s8K ctx
dense
GPT-OSS 20B21B18.7 GB14 tok/s6K ctx
moe
CodeGeeX 4 9B9B9.6 GB16 tok/s131K ctx
dense
Gemma 3 12B12B15.7 GB8 tok/s21K ctx
dense
Gemma 4 E4B8B9.7 GB13 tok/s111K ctx
dense
Granite Code 8B8B10.3 GB18 tok/s8K ctx
dense
Qwen 2.5 Coder 3B3B7.5 GB42 tok/s87K ctx
dense
Granite 4.1 8B8B10.8 GB18 tok/s58K ctx
dense
DevStral 7B7B9.7 GB20 tok/s8K ctx
dense
Pixtral 12B12B13.3 GB11 tok/s42K ctx
dense
mxbai Embed Large0.34B5.6 GB5 tok/s1K ctx
dense
Qwen 2.5 7B7B8.6 GB20 tok/s131K ctx
dense
Codestral Mamba 7B7B8.3 GB23 tok/s262K ctx
state-space
Snowflake Arctic Embed L0.34B5.6 GB5 tok/s1K ctx
dense
StarCoder 7B7B15.1 GB19 tok/s8K ctx
dense
CodeLlama 7B Instruct7B15.6 GB19 tok/s16K ctx
dense
Magistral Small 250724B20.6 GB7 tok/s4K ctx
dense
Devstral Small 2 24B Instruct24B20.6 GB7 tok/s4K ctx
dense
DeepSeek R1 Distill 14B14B15.0 GB10 tok/s29K ctx
dense
BGE Large EN v1.50.34B5.6 GB5 tok/s1K ctx
dense
Solar 7B7B10.7 GB20 tok/s8K ctx
dense
Llama 3.1 8B8B10.3 GB18 tok/s73K ctx
dense
Codestral 2 25.0822B19.4 GB8 tok/s4K ctx
dense
Phi 3 Mini 3.8B3.8B11.7 GB34 tok/s31K ctx
dense
B
B Tier -- Good Fit (42 models)
Solid performance, may need quantization or reduced context.
WizardMath 7B7B9.7 GB20 tok/s4K ctx
dense
OLMo 2 7B7B9.7 GB20 tok/s4K ctx
dense
GLM-4 9B9B9.6 GB16 tok/s128K ctx
dense
Devstral Small 1.124B20.6 GB7 tok/s4K ctx
dense
InternLM 7B7B15.6 GB19 tok/s8K ctx
dense
InternLM Chat 7B7B15.6 GB19 tok/s8K ctx
dense
Ministral 3 3B3B6.1 GB42 tok/s261K ctx
multimodal
Gemma 3 4B4B8.0 GB26 tok/s87K ctx
dense
Phi 3.5 Mini 4B4B11.8 GB33 tok/s31K ctx
dense
Gemma 4 E2B5.1B7.1 GB21 tok/s128K ctx
dense
Phi 4 Mini 4B4B7.4 GB35 tok/s124K ctx
dense
Qwen 3.5 2B2B6.4 GB28 tok/s118K ctx
dense
Nomic Embed Text v1.50.14B4.3 GB2 tok/s8K ctx
dense
Qwen 2.5 Coder 7B7B8.6 GB20 tok/s131K ctx
dense
Qwen 2.5 3B3B7.5 GB42 tok/s87K ctx
dense
LLaVA 1.5 7B7B15.6 GB19 tok/s4K ctx
dense
Granite Code 20B20B18.9 GB9 tok/s8K ctx
dense
DeepSeek R1 Distill 8B8B10.3 GB18 tok/s33K ctx
dense
Granite Code 3B3B7.8 GB42 tok/s8K ctx
dense
Samantha 7B7B9.7 GB20 tok/s4K ctx
dense
Qwen 3 1.7B1.7B6.2 GB24 tok/s33K ctx
dense
MPT-7B-Instruct7B15.6 GB19 tok/s8K ctx
dense
Baichuan 7B7B15.6 GB19 tok/s8K ctx
dense
Llama 3.2 11B Vision11B12.2 GB12 tok/s16K ctx
vision
Gemma 2 9B9B14.1 GB12 tok/s8K ctx
dense
DeepSeek R1 Distill 7B7B8.6 GB20 tok/s33K ctx
dense
Granite 4.1 3B3B6.5 GB42 tok/s131K ctx
dense
Mistral Small 3.2 24B24B20.6 GB7 tok/s4K ctx
vision
LFM2 24B24B20.6 GB7 tok/s4K ctx
dense
Falcon 7B Instruct7B7.9 GB21 tok/s8K ctx
dense
Mistral Nemo 12B12B13.3 GB11 tok/s42K ctx
dense
All MiniLM L6 v20.02B3.8 GB2 tok/s0K ctx
dense
Qwen 2.5 Coder 1.5B1.5B4.8 GB21 tok/s33K ctx
dense
Mistral Small 24B24B20.6 GB7 tok/s4K ctx
dense
Yi Coder 9B9B10.4 GB16 tok/s91K ctx
dense
Mistral 7B Instruct v0.37B9.7 GB20 tok/s8K ctx
dense
Mistral Small 3.1 24B24B20.6 GB7 tok/s4K ctx
dense
Qwen 2.5 Coder 14B14B15.0 GB10 tok/s29K ctx
dense
Llama 3.2 3B3B7.0 GB42 tok/s112K ctx
dense
Ministral 8B8B10.6 GB18 tok/s65K ctx
dense
Phi 3 Medium 14B14B15.1 GB10 tok/s28K ctx
dense
SmolLM3 3B3B7.3 GB42 tok/s98K ctx
dense
C
C Tier -- Tight Fit (169 models)
Runs with offloading or significant compromises.
Granite 3.1 8B8B10.3 GB22 tok/s73K ctx
state-space
DeepSeek R1 1.5B1.5B4.8 GB21 tok/s33K ctx
dense
Yi 1.5 9B9B10.4 GB16 tok/s4K ctx
dense
TinyLlama 1.1B1.1B4.5 GB15 tok/s4K ctx
dense
Qwen 2.5 Math 7B7B8.6 GB20 tok/s4K ctx
dense
Qwen 2.5 1.5B1.5B4.8 GB21 tok/s131K ctx
dense
OpenChat 7B7B9.7 GB20 tok/s8K ctx
dense
Aya Expanse 8B8B10.3 GB18 tok/s8K ctx
dense
Gemma 2 2B2B6.3 GB28 tok/s8K ctx
dense
gemma 3 12b it
C5012B12.2 GB10 tok/s74K ctx
dense
Dolphin 2.9 8B8B10.3 GB18 tok/s33K ctx
dense
OpenHermes 2.5 7B7B9.7 GB20 tok/s8K ctx
dense
DeepSeek R1 Distill Qwen 14B14B13.7 GB9 tok/s51K ctx
dense
Nemotron Mini 4B4B7.9 GB38 tok/s4K ctx
dense
Zephyr 7B Beta7B9.7 GB20 tok/s33K ctx
dense
Gemma 3 1B1B4.5 GB14 tok/s33K ctx
dense
NousResearch Hermes 4 14B
C4914B13.7 GB9 tok/s51K ctx
dense
Vicuna 7B7B15.6 GB19 tok/s4K ctx
dense
Starling LM 7B7B9.7 GB20 tok/s8K ctx
dense
SOLAR 10.7B Instruct v1.0 uncensored
C4910.7B11.3 GB12 tok/s93K ctx
dense
Nous Hermes 2 SOLAR 10.7B
C4910.7B11.3 GB12 tok/s93K ctx
dense
GGUF SOLARized GraniStral 14B 2102 YeAM HCT 32QKV
C4914B13.7 GB9 tok/s51K ctx
dense
internlm JanusCoder 14B
C4914B13.7 GB9 tok/s51K ctx
dense
GGUF SOLARized GraniStral 14B 1902 YeAM HCT
C4914B13.7 GB9 tok/s51K ctx
dense
solar finalised finetuned Model 10.7B i1
C4910.7B11.3 GB12 tok/s93K ctx
dense
SOLAR 10.7B v1.0
C4910.7B11.3 GB12 tok/s93K ctx
dense
Qwen3.5 9B9B10.0 GB15 tok/s126K ctx
dense
Qwen3.5 9B Uncensored HauhauCS Aggressive
C499B10.0 GB15 tok/s126K ctx
dense
Qwen3.5 9B
C499B10.0 GB15 tok/s126K ctx
dense
HelpingAI2.5 10B i1
C4810B10.8 GB13 tok/s105K ctx
dense
Yi Coder 9B Chat
C489B10.0 GB15 tok/s126K ctx
dense
glm 4 9b chat 1m
C489B10.0 GB15 tok/s126K ctx
dense
Helply 10.2b chat i1
C4810.2B10.9 GB12 tok/s101K ctx
dense
Neural Chat 7B7B9.7 GB20 tok/s8K ctx
dense
Yi 1.5 9B Chat
C489B10.0 GB15 tok/s126K ctx
dense
Meta Llama 3.1 8B Instruct
C488B9.3 GB16 tok/s152K ctx
dense
llava llama 3 8b v1 1
C488B9.3 GB16 tok/s152K ctx
dense
DeepSeek R1 0528 Qwen3 8B8B9.3 GB16 tok/s152K ctx
dense
Meta Llama 3 8B Instruct
C488B9.3 GB16 tok/s152K ctx
dense
DeepSeek R1 Distill Llama 8B8B9.3 GB16 tok/s152K ctx
dense
Dolphin3.0 Llama3.1 8B
C488B9.3 GB16 tok/s152K ctx
dense
Llama 3 8B Instruct 32k v0.1
C488B9.3 GB16 tok/s152K ctx
dense
Meta Llama 3.1 8B Instruct
C488B9.3 GB16 tok/s152K ctx
dense
vntl llama3 8b v2
C488B9.3 GB16 tok/s152K ctx
dense
Yi 9B Coder i1
C489B10.0 GB15 tok/s126K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C488B9.3 GB16 tok/s152K ctx
dense
falcon mamba 7b instruct Q4 K M
C487B8.6 GB23 tok/s186K ctx
dense
HelpingAI2 9B
C489B10.0 GB15 tok/s126K ctx
dense
DeepSeek R1 0528 Qwen3 8B
C488B9.3 GB16 tok/s152K ctx
dense
Qwen3 8B DeepSeek v3.2 Speciale Distill
C488B9.3 GB16 tok/s152K ctx
dense
HelpingAI2 9B i1
C489B10.0 GB15 tok/s126K ctx
dense
Hermes 3 Llama 3.1 8B
C488B9.3 GB16 tok/s152K ctx
dense
HelpingAI 9B 200k i1
C489B10.0 GB15 tok/s126K ctx
dense
HelpingAI 9B i1
C489B10.0 GB15 tok/s126K ctx
dense
Hermes 2 Pro Llama 3 8B
C488B9.3 GB16 tok/s152K ctx
dense
Dolphin3.0 Llama3.1 8B
C488B9.3 GB16 tok/s152K ctx
dense
granite 8b code instruct 4k
C488B9.3 GB16 tok/s152K ctx
dense
Mamba Codestral 7B v0.1
C487B8.6 GB23 tok/s186K ctx
dense
dolphin 2.9.4 llama3.1 8b
C488B9.3 GB16 tok/s152K ctx
dense
Llama 2 7B Chat
C487B8.6 GB19 tok/s186K ctx
dense
openchat 3.6 8b 20240522 IMat
C488B9.3 GB16 tok/s152K ctx
dense
Mistral 7B Instruct v0.2
C487B8.6 GB19 tok/s186K ctx
dense
Qwen 2.5 Coder 0.5B0.5B4.0 GB7 tok/s131K ctx
dense
aya expanse 8b
C488B9.3 GB16 tok/s152K ctx
dense
Mistral 7B Instruct v0.3
C487B8.6 GB19 tok/s186K ctx
dense
StarCoder2 15B15B15.5 GB8 tok/s16K ctx
dense
dolphin v2 8b abliterated i1
C488B9.3 GB16 tok/s152K ctx
dense
EXAONE 3.5 7.8B Instruct
C487.8B9.2 GB17 tok/s158K ctx
dense
EXAONE 3.5 7.8B Instruct
C487.8B9.2 GB17 tok/s158K ctx
dense
aya expanse 8b orthogonal heretic i1
C488B9.3 GB16 tok/s152K ctx
dense
EXAONE 3.5 7.8B Instruct i1
C487.8B9.2 GB17 tok/s158K ctx
dense
exaone 3.0 7.8b it
C487.8B9.2 GB17 tok/s158K ctx
dense
aya expanse 8b orthogonal heretic
C478B9.3 GB16 tok/s152K ctx
dense
Mistral 7B Instruct v0.3
C477B8.6 GB19 tok/s186K ctx
dense
Hermes 2 Pro Mistral 7B
C477B8.6 GB19 tok/s186K ctx
dense
internlm2 5 1 8b chat i1
C478B9.3 GB16 tok/s152K ctx
dense
internlm3 8b instruct abliterated i1
C478B9.3 GB16 tok/s152K ctx
dense
zephyr 7B beta
C477B8.6 GB19 tok/s186K ctx
dense
Yi 1.5 6B6B8.1 GB24 tok/s4K ctx
dense
Nous Hermes 2 Mistral 7B DPO
C477B8.6 GB19 tok/s186K ctx
dense
Falcon H1R 7B
C477B8.6 GB19 tok/s186K ctx
dense
zephyr 7B alpha
C477B8.6 GB19 tok/s186K ctx
dense
speechless zephyr code functionary 7b
C477B8.6 GB19 tok/s186K ctx
dense
StarCoder2 7B
C477B8.6 GB19 tok/s186K ctx
dense
japanese stablelm instruct gamma 7B
C477B8.6 GB19 tok/s186K ctx
dense
Falcon H1 7B Instruct
C477B8.6 GB19 tok/s186K ctx
dense
Falcon H1R 7B7B8.6 GB19 tok/s186K ctx
dense
DeepSeek LLM 7B7B15.1 GB19 tok/s4K ctx
dense
Yi 1.5 6B Chat
C476B7.9 GB22 tok/s230K ctx
dense
stabilityai japanese stablelm base gamma 7b
C477B8.6 GB19 tok/s186K ctx
dense
stablelm 2 zephyr 1 6b
C476B7.9 GB22 tok/s230K ctx
dense
baichuan2 7b chat
C477B8.6 GB19 tok/s186K ctx
dense
DiscoPOP zephyr 7b gemma
C477B8.6 GB19 tok/s186K ctx
dense
starcoder2 7b
C477B8.6 GB19 tok/s186K ctx
dense
zephyr 7b beta Mistral 7B Instruct v0.2
C477B8.6 GB19 tok/s186K ctx
dense
OpenChat 3.5 7B Qwen v2.0 i1
C477B8.6 GB19 tok/s186K ctx
dense
OpenChat 3.5 7B Starling v2.0 i1
C477B8.6 GB19 tok/s186K ctx
dense
internlm2 math plus 7b IMat
C477B8.6 GB19 tok/s186K ctx
dense
CodeNinja 1.0 OpenChat 7B i1
C477B8.6 GB19 tok/s186K ctx
dense
internlm2 5 7b chat i1
C477B8.6 GB19 tok/s186K ctx
dense
OpenSafetyLab MD Judge v0 2 internlm2 7b
C477B8.6 GB19 tok/s186K ctx
dense
MD Judge v0 2 internlm2 7b i1
C477B8.6 GB19 tok/s186K ctx
dense
zephyr 7b gemma sft african ultrachat 100k
C477B8.6 GB19 tok/s186K ctx
dense
jointpreferences mistral 7b sft helpful
C477B8.6 GB19 tok/s186K ctx
dense
zephyr 7b dpo full i1
C477B8.6 GB19 tok/s186K ctx
dense
blossom v3 baichuan2 7b i1
C477B8.6 GB19 tok/s186K ctx
dense
Yi 1.5 6B Chat
C476B7.9 GB22 tok/s230K ctx
dense
blossom v1 baichuan 7b i1
C477B8.6 GB19 tok/s186K ctx
dense
stablelm 2 1 6b chat imatrix
C476B7.9 GB22 tok/s230K ctx
dense
Qwen3.5 4B4B6.4 GB33 tok/s387K ctx
dense
Llama 3.2 3B Instruct
C473B6.0 GB38 tok/s529K ctx
dense
Qwen2.5 3B Instruct3B5.7 GB42 tok/s544K ctx
dense
gemma 3 4b it
C474B6.4 GB33 tok/s387K ctx
dense
gemma 3 4b it
C474B6.4 GB33 tok/s387K ctx
dense
HelpingAI2 6B
C476B7.9 GB22 tok/s230K ctx
dense
StarCoder2 7B7B8.3 GB20 tok/s16K ctx
dense
HelpingAI2 6B i1
C476B7.9 GB22 tok/s230K ctx
dense
Llama 3.2 3B Instruct
C473B5.7 GB42 tok/s544K ctx
dense
Ministral 3 3B Instruct 2512
C463B5.7 GB42 tok/s544K ctx
dense
Qwen 3 0.6B0.6B4.7 GB8 tok/s33K ctx
dense
Phi 4 reasoning vision 15B
C4615B14.4 GB9 tok/s42K ctx
dense
HELVETE 3B
C463B5.7 GB42 tok/s544K ctx
dense
StarCoder2 15B
C4615B14.4 GB9 tok/s42K ctx
dense
Hermes 3 Llama 3.2 3B
C463B5.7 GB42 tok/s544K ctx
dense
Codestral 22B22B19.4 GB8 tok/s4K ctx
dense
stablelm zephyr 3b
C463B5.7 GB42 tok/s544K ctx
dense
StarCoder2 3B
C463B5.7 GB42 tok/s544K ctx
dense
Llama 3.2 1B1B4.6 GB14 tok/s128K ctx
dense
AI21 Jamba Reasoning 3B
C463B5.7 GB42 tok/s544K ctx
dense
HelpingAI2.5 5B i1
C465B7.1 GB26 tok/s293K ctx
dense
starcoder2 15b instruct v0.1
C4615B14.4 GB9 tok/s42K ctx
dense
starcoder2 15b i1
C4615B14.4 GB9 tok/s42K ctx
dense
Codestral RAG 19B Pruned i1
C4619B17.3 GB9 tok/s16K ctx
dense
stablelm 3b 4e1t
C463B5.7 GB42 tok/s544K ctx
dense
ai21labs AI21 Jamba Reasoning 3B
C463B5.7 GB42 tok/s544K ctx
dense
starcoder2 15b instruct v0.1
C4615B14.4 GB9 tok/s42K ctx
dense
ai21labs AI21 Jamba2 3B
C463B5.7 GB42 tok/s544K ctx
dense
internlm2 5 20b chat
C4620B18.0 GB9 tok/s11K ctx
dense
HelpingAI 15B i1
C4615B14.4 GB9 tok/s42K ctx
dense
HelpingAI 3B hindi i1
C463B5.7 GB42 tok/s544K ctx
dense
internlm2 limarp chat 20b
C4620B18.0 GB9 tok/s11K ctx
dense
AI21 Jamba2 3B
C463B5.7 GB42 tok/s544K ctx
dense
HelpingAI 3B hindi
C463B5.7 GB42 tok/s544K ctx
dense
AI21 Jamba2 3B i1
C463B5.7 GB42 tok/s544K ctx
dense
internlm2 math plus 20b i1
C4620B18.0 GB9 tok/s11K ctx
dense
gemma 2 2b it
C452B5.4 GB28 tok/s829K ctx
dense
StarCoder2 3B3B5.8 GB42 tok/s16K ctx
dense
EXAONE 3.5 2.4B Instruct
C452.4B5.2 GB34 tok/s701K ctx
dense
gemma 2b2B4.9 GB28 tok/s858K ctx
dense
Gemmasutra Mini 2B v1
C442B4.9 GB28 tok/s858K ctx
dense
gemma 2 2b it
C442B4.9 GB28 tok/s858K ctx
dense
Qwen 3.5 0.6B0.6B4.7 GB8 tok/s131K ctx
dense
Qwen2.5 1.5B Instruct1.5B4.6 GB21 tok/s1172K ctx
dense
DeepSeek R1 Distill Qwen 1.5B1.5B4.6 GB21 tok/s1172K ctx
dense
Yi Coder 1.5B Chat
C431.5B4.6 GB21 tok/s1172K ctx
dense
stablelm 2 zephyr 1.6b
C431.6B4.7 GB22 tok/s1093K ctx
dense
logos16v2 stablelm2 1.6b i1
C431.6B4.7 GB22 tok/s1093K ctx
dense
Yi Coder 1.5B
C431.5B4.6 GB21 tok/s1172K ctx
dense
Falcon H1 1.5B Instruct1.5B4.6 GB21 tok/s1172K ctx
dense
TinyLlama 1.1B Chat v1.0
C421.1B4.3 GB15 tok/s1628K ctx
dense
EXAONE 4.0 1.2B
C421.2B4.4 GB17 tok/s1485K ctx
dense
Llama 3.2 1B Instruct Q8 0
C421B4.4 GB14 tok/s1771K ctx
dense
TinyLlama 1.1B Chat v0.3
C411.1B4.3 GB15 tok/s1628K ctx
dense
TinyLlama 1.1B Chat v0.61.1B4.3 GB15 tok/s1628K ctx
dense
Qwen 2.5 0.5B0.5B4.0 GB7 tok/s131K ctx
dense
Llama 3.2 1B Instruct
C411B4.2 GB14 tok/s1799K ctx
dense
gemma 3 1b it
C411B4.2 GB14 tok/s1799K ctx
dense
TinyLlama 1.1B Chat v1.0 imatrix
C411.1B4.3 GB15 tok/s1628K ctx
dense
Falcon3 1B Instruct abliterated
C411B4.2 GB14 tok/s1799K ctx
dense
F
F Tier -- Can't Run (117 models)
Too heavy for this hardware.
SmolVLM 500M Instruct
D390.5B4.0 GB7 tok/s2140K ctx
dense
embeddinggemma 300M
D370.3B3.8 GB4 tok/s2167K ctx
dense
Codestral 22B v0.1
D3622B19.5 GB7 tok/s4K ctx
dense
Codestral 22B v0.1
D3622B19.5 GB7 tok/s4K ctx
dense
Codestral 22B v0.1
D3622B19.5 GB7 tok/s4K ctx
dense
Codestral 21B Pruned i1
D3621B18.8 GB8 tok/s6K ctx
dense
Codestral 22B v0.1 IMat
D3622B19.5 GB7 tok/s4K ctx
dense
Codestral 22B v0.1 i1
D3522B19.5 GB7 tok/s4K ctx
dense
granite embedding 107m multilingual
D350.11B3.7 GB2 tok/s2196K ctx
dense
Falcon H1 Tiny 90M Instruct
D350.09B3.6 GB2 tok/s2197K ctx
dense
Llama 3.1 70B70B51.1 GB2 tok/s4K ctx
dense
Llama 3.3 70B70B51.1 GB2 tok/s4K ctx
dense
Qwen 2.5 32B32B26.9 GB5 tok/s4K ctx
dense
Qwen 2.5 72B72B52.3 GB2 tok/s4K ctx
dense
Qwen 2.5 Coder 32B32B26.9 GB5 tok/s4K ctx
dense
Qwen 3 32B32B26.9 GB5 tok/s4K ctx
dense
Qwen 3 235B A22B235B149.7 GB2 tok/s4K ctx
moe
Qwen 3 30B A3B30.5B23.6 GB9 tok/s4K ctx
moe
Qwen3-Coder 30B A3B Instruct30.5B23.6 GB9 tok/s4K ctx
moe
Qwen3-Coder-Next80B53.8 GB3 tok/s4K ctx
moe
Qwen3-Coder 480B A35B Instruct480B299.2 GB2 tok/s4K ctx
moe
Qwen3-VL 30B A3B Instruct30B23.3 GB9 tok/s4K ctx
moe
Qwen 3.5 27B27B23.1 GB6 tok/s4K ctx
dense
Qwen 3.5 35B A3B35B26.3 GB8 tok/s4K ctx
moe
Qwen 3.6 35B A3B35B28.9 GB6 tok/s4K ctx
moe
Qwen 3.6 27B27B20.9 GB5 tok/s4K ctx
dense
Qwen 3.5 122B A10B122B80.4 GB2 tok/s4K ctx
moe
Qwen 3.5 397B A17B397B248.5 GB2 tok/s4K ctx
moe
Gemma 2 27B27B31.2 GB3 tok/s4K ctx
dense
Gemma 3 27B27B31.2 GB3 tok/s4K ctx
dense
Mixtral 8x7B47B34.1 GB3 tok/s4K ctx
moe
Mixtral 8x22B141B92.9 GB2 tok/s4K ctx
moe
DeepSeek R1 Distill 32B32B26.9 GB5 tok/s4K ctx
dense
Granite 4.1 30B30B25.7 GB5 tok/s4K ctx
dense
Command R 35B35B27.3 GB4 tok/s4K ctx
dense
Falcon 40B Instruct40B34.1 GB3 tok/s4K ctx
dense
WizardLM 13B13B23.6 GB6 tok/s8K ctx
dense
StarCoder 15B15B28.9 GB4 tok/s4K ctx
dense
CodeLlama 13B Instruct13B23.6 GB6 tok/s8K ctx
dense
MPT-30B-Instruct30B48.5 GB3 tok/s4K ctx
dense
Devstral 2 123B Instruct123B83.9 GB2 tok/s4K ctx
dense
Mistral Small 4 119B119B81.5 GB3 tok/s4K ctx
moe
Leanstral 119B A6B119B84.9 GB2 tok/s4K ctx
moe
Mistral Large 3675B420.6 GB2 tok/s4K ctx
moe
GLM-5744B476.4 GB2 tok/s4K ctx
moe
GLM-5.1754B482.5 GB2 tok/s4K ctx
moe
Kimi K2.51000B620.9 GB2 tok/s4K ctx
moe
Kimi K2.61000B620.9 GB2 tok/s4K ctx
moe
Kimi Linear 48B A3B48B33.7 GB2 tok/s4K ctx
linear
Nous Hermes 1.09B21.2 GB11 tok/s11K ctx
dense
Nous Dolphin 13B13B25.1 GB5 tok/s6K ctx
dense
Baichuan 13B13B25.1 GB5 tok/s6K ctx
dense
Cerebras-GPT 13B13B22.6 GB6 tok/s7K ctx
dense
StableLM 2 12B12B24.3 GB5 tok/s4K ctx
dense
InternLM 20B20B36.2 GB4 tok/s4K ctx
dense
Yi 34B Chat34B27.9 GB4 tok/s4K ctx
dense
DeepSeek V3 671B671B472.4 GB2 tok/s4K ctx
moe
DeepSeek V3.1 671B671B472.4 GB2 tok/s4K ctx
moe
DeepSeek V4 Pro1600B867.4 GB2 tok/s4K ctx
moe
DeepSeek V4 Flash284B162.8 GB2 tok/s4K ctx
moe
DeepSeek R1 671B671B472.4 GB2 tok/s4K ctx
moe
DeepSeek Coder V2 236B236B206.0 GB2 tok/s4K ctx
moe
DeepSeek V2.5 236B236B206.0 GB2 tok/s4K ctx
moe
DeepSeek LLM 67B67B50.2 GB2 tok/s4K ctx
dense
Llama 4 Scout 17B 16E109B72.9 GB2 tok/s4K ctx
moe
Llama 4 Maverick 17B 128E400B250.4 GB2 tok/s4K ctx
moe
Command R+ 104B104B70.3 GB2 tok/s4K ctx
dense
Command A 111B111B75.1 GB2 tok/s4K ctx
dense
Yi 1.5 34B34B27.9 GB4 tok/s4K ctx
dense
Pixtral Large 124B124B84.5 GB2 tok/s4K ctx
dense
Qwen 2.5 VL 72B72B52.3 GB2 tok/s4K ctx
dense
Qwen 2.5 Math 72B72B52.3 GB2 tok/s4K ctx
dense
Nemotron 70B70B51.1 GB2 tok/s4K ctx
dense
LLaVA 1.6 13B13B23.6 GB6 tok/s4K ctx
dense
Vicuna 13B13B23.6 GB6 tok/s4K ctx
dense
OLMo 2 32B32B26.9 GB5 tok/s4K ctx
dense
Nemotron 3 Nano 30B30B24.2 GB6 tok/s4K ctx
dense
EXAONE 4.0 32B32B26.9 GB5 tok/s4K ctx
dense
Llama 3.1 405B405B258.2 GB2 tok/s4K ctx
dense
DeepSeek R1 Distill 70B70B51.1 GB2 tok/s4K ctx
dense
Granite Code 34B34B27.9 GB4 tok/s4K ctx
dense
GPT-OSS 120B117B79.7 GB2 tok/s4K ctx
dense
Aya Expanse 32B32B25.5 GB5 tok/s4K ctx
dense
Gemma 4 26B A4B25.2B22.5 GB10 tok/s4K ctx
moe
Gemma 4 31B30.7B36.9 GB3 tok/s4K ctx
dense
DeepSeek V3.2671B413.3 GB2 tok/s4K ctx
moe
Nemotron Cascade 2 30B A3B30B24.7 GB8 tok/s4K ctx
moe
MiniMax M2.7
F0230B147.6 GB2 tok/s4K ctx
moe
Solar Open 100B
F0100B76.2 GB2 tok/s4K ctx
dense
Baichuan M2 32B Q4 K M
F032B26.8 GB5 tok/s4K ctx
dense
baichuan inc Baichuan M2 32B
F032B26.8 GB5 tok/s4K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
F024B20.9 GB7 tok/s4K ctx
dense
cognitivecomputations Dolphin3.0 R1 Mistral 24B
F024B20.9 GB7 tok/s4K ctx
dense
Qwen3 48B A4B Savant Commander Distill 12X Closed Open Heretic Uncensored
F048B38.4 GB2 tok/s4K ctx
dense
EXAONE 4.0 32B
F032B26.8 GB5 tok/s4K ctx
dense
K EXAONE 236B A23B
F0236B175.1 GB2 tok/s4K ctx
dense
Qwen3.5 35B A3B
F035B28.9 GB4 tok/s4K ctx
dense
gemma 3 27b it
F027B23.1 GB6 tok/s4K ctx
dense
Llama 3.3 70B Instruct
F070B54.4 GB2 tok/s4K ctx
dense
Mistral Small 24B Instruct 2501
F024B20.9 GB7 tok/s4K ctx
dense
mistral small 3.1 24b instruct 2503 hf
F024B20.9 GB7 tok/s4K ctx
dense
aya expanse 32b heretic MPOA i1
F032B26.8 GB5 tok/s4K ctx
dense
Baichuan M3 235B
F0235B174.4 GB2 tok/s4K ctx
dense
Baichuan M3 235B i1
F0235B174.4 GB2 tok/s4K ctx
dense
BaichuanMed OCR 72B i1
F072B55.8 GB2 tok/s4K ctx
dense
Dolphin Mistral GLM 4.7 Flash 24B Venice Edition Thinking Uncensored i1
F024B20.9 GB7 tok/s4K ctx
dense
Solar Open 100B i1
F0100B76.2 GB2 tok/s4K ctx
dense
Solar Open 69B REAP i1
F069B53.7 GB2 tok/s4K ctx
dense
Hermes 4.3 36B
F036B29.7 GB4 tok/s4K ctx
dense
stabilityai japanese stablelm instruct beta 70b
F070B54.4 GB2 tok/s4K ctx
dense
gemma 3 27b it27B23.1 GB6 tok/s4K ctx
dense
Mistral Small 3.2 24B Instruct 250624B20.9 GB7 tok/s4K ctx
dense
Qwen3.5 122B A10B122B77.6 GB2 tok/s4K ctx
dense
Qwen3.5 27B27B23.1 GB6 tok/s4K ctx
dense
Qwen3.5 35B A3B35B28.9 GB4 tok/s4K ctx
dense
Qwen3.5 397B A17B397B292.2 GB2 tok/s4K ctx
dense
cognitivecomputations Dolphin Mistral 24B Venice Edition
F024B20.9 GB7 tok/s4K ctx
dense
Compare with...
See How Other Hardware Stacks Up
Check tier lists for similar GPUs and Apple Silicon configs