Hardware / Apple /Apple M3 Ultra (80-core GPU)

Apple · Apple Silicon

Apple M3 Ultra (80-core GPU)

for example Mac Studio (2025). Memory is a build option on these, so not every one of them is sold with 512 GB.

Runs 198 models fully in memory, 12 more with CPU offload. 2 of those sit close enough to 512 GB that our size estimate is what decides them.

Memory

512 GB

Bandwidth

819 GB/s

Runs

198

01

Specifications

Memory

512 GB

Decides which models fit at all.

Memory bandwidth

819 GB/s

How fast it can read the model. Sets the speed of the reply.

Compute

56.8 TFLOPS

Trillions of half-precision sums a second. Sets how fast it reads a long prompt, not how fast it replies.

Power draw

140 W

At full load, for sizing a power supply.

Released

Mar 12, 2025

Architecture

M3

02

What it runs

Best quant per model. Text models first, then speech, then what does not fit.

What is quantisation? →

ModelParamsContextSpeedVerdict

Qwen3.6 35B A3B36B262K228 tok/sFits in memory Gemma 4 26B A4B25.8B262K171 tok/sFits in memory Nemotron 3 Nano 30B A3B31.6B262K268 tok/sFits in memory Ministral 3 3B 25123.8B131K249 tok/sFits in memory gpt-oss-safeguard-20b21.5B131K223 tok/sFits in memory Granite 4.0 Micro3.2B66K296 tok/sFits in memory Qwen3 VL 30B A3B Thinking31.1B262K153 tok/sFits in memory Qwen3 Next 80B A3B Thinking81.3B262K153 tok/sFits in memory gpt-oss-120b120B131K134 tok/sFits in memory Llama 3.2 3B Instruct3.2B131K252 tok/sFits in memory Qwen3.5-35B-A3B36B262K268 tok/sFits in memory Qwen3 VL 30B A3B Instruct31.1B262K153 tok/sFits in memory Qwen3 Next 80B A3B Instruct81.3B262K228 tok/sFits in memory Qwen3 30B A3B Thinking 250730.5B66K228 tok/sFits in memory Qwen3 Coder 30B A3B Instruct30.5B262K153 tok/sFits in memory Qwen3 30B A3B Instruct 250730.5B262K153 tok/sFits in memory Gemma 3 4B4.3B131K126 tok/sFits in memory Llama 3.1 8B Instruct8B131K118 tok/sFits in memory Nemotron 3.5 Content Safety4.3B131K188 tok/sFits in memory Qwen3 8B8.2B131K115 tok/sFits in memory UI-TARS 7B8.3B66K114 tok/sFits in memory Granite 4.1 8B8.8B131K108 tok/sFits in memory Ministral 3 8B 25128.9B262K106 tok/sFits in memory Qwen3 VL 8B Thinking8.8B131K108 tok/sFits in memory Gemma 3n 4B7.8B33K121 tok/sFits in memory Llama 3.2 1B Instruct1.2B33K672 tok/sFits in memory Qwen3 VL 8B Instruct8.8B262K108 tok/sFits in memory Qwen2.5 7B Instruct7.6B33K124 tok/sFits in memory Granite 4.2 8B8.8B131K108 tok/sFits in memory Reka Edge7.1B16K133 tok/sFits in memory Qwen3.5-9B9.7B262K98 tok/sFits in memory Llama 3 8B Lunaris8B8K118 tok/sFits in memory Hy-MT2-1.8B2B8K343 tok/sFits in memory Hy-MT2-30B-A3B30.1B8K228 tok/sFits in memory MiniMax M2.7229B131K80 tok/sFits in memory Qwen3.5-122B-A10B125B262K80 tok/sFits in memory Llama Guard 4 12B12B262K79 tok/sFits in memory MiniMax M2.5229B131K80 tok/sFits in memory MiniMax M2.1229B131K80 tok/sFits in memory MiniMax M2229B131K80 tok/sFits in memory Gemma 3 12B12.2B131K78 tok/sFits in memory Rocinante 12B12.2B66K78 tok/sFits in memory Mistral Nemo12.2B131K78 tok/sFits in memory Ministral 3 14B 251213.9B262K68 tok/sFits in memory GLM 4.5 Air111B131K67 tok/sFits in memory Qwen3 14B14.8B131K64 tok/sFits in memory UnslopNemo 12B12.2B33K78 tok/sFits in memory Hy-MT2-7B8B8K101 tok/sFits in memory Hunyuan A13B Instruct80.4B131K62 tok/sFits in memory Qwen3.5 397B A17B403B262K47 tok/sFits in memory Reka Flash 320.9B66K45 tok/sFits in memory Cydonia 24B V4.123.6B131K40 tok/sFits in memory Uncensored24B66K39 tok/sFits in memory Mistral Small 3.2 24B24B131K39 tok/sFits in memory Phi 414.7B16K64 tok/sFits in memory Mistral Small 3.1 24B24B66K39 tok/sFits in memory Qwen3.5-27B27.8B262K34 tok/sFits in memory Qwen3 VL 235B A22B Thinking236B131K37 tok/sFits in memory Gemma 3 27B27.4B262K35 tok/sFits in memory MythoMax 13B13B8K73 tok/sFits in memory Qwen3.6 27B27.8B262K34 tok/sFits in memory Qwen3 VL 235B A22B Instruct236B262K37 tok/sFits in memory Qwen3 235B A22B Thinking 2507235B262K37 tok/sFits in memory Qwen3 235B A22B Instruct 2507235B262K37 tok/sFits in memory Qwen3.8 27B27.8B262K34 tok/sFits in memory Ternary Bonsai 2 27B27B262K35 tok/sFits in memory Gemma 4 31B31.3B262K30 tok/sFits in memory Muse Glimmer 30B29.8B131K32 tok/sFits in memory Olmo 3 32B Think32.2B66K29 tok/sFits in memory Qwen3 VL 32B Instruct33.4B131K28 tok/sFits in memory Qwen3 32B32.8B131K29 tok/sFits in memory Mistral Small 323.6B33K40 tok/sFits in memory Nex-N2-Mini35.1B262K23 tok/sFits in memory GLM 4.7 Flash31.2B131K22 tok/sFits in memory ReMM SLERP 13B13B4K73 tok/sFits in memory Laguna XS 2.133.4B262K24 tok/sFits in memory GLM 4.5358B131K25 tok/sFits in memory Qwen3 Coder 480B A35B480B262K23 tok/sFits in memory Nemotron 3.5 Lightning31.6B262K25 tok/sFits in memory Nex-N2.5-Mini35.1B262K23 tok/sFits in memory Mixtral 8x22B Instruct141B66K21 tok/sFits in memory DeepSeek V4 Flash291B262K19 tok/sFits in memory DeepSeek V4 Flash Vision Exp305B262K19 tok/sFits in memory Voxtral Small 24B 250724.3B16K39 tok/sFits in memory ERNIE 4.5 VL 424B A47B424B66K15 tok/sFits in memory Qwen2.5 Coder 32B Instruct32.8B33K29 tok/sFits in memory Qwen3 Coder Next79.7B262K10 tok/sFits in memory Hermes 4 70B70.6B131K13 tok/sFits in memory Skyfall 36B V236.9B33K26 tok/sFits in memory Qwen2.5 VL 72B Instruct73.4B66K13 tok/sFits in memory Llama 3.3 Euryale 70B70.6B131K11 tok/sFits in memory Llama 3.1 Euryale 70B v2.270.6B131K13 tok/sFits in memory Hermes 3 70B Instruct70.6B131K11 tok/sFits in memory Llama 3.1 70B Instruct70.6B131K13 tok/sFits in memory Llama 3.3 70B Instruct70.6B131K13 tok/sFits in memory Laguna S 2.1118B262K7 tok/sFits in memory Mistral Medium 3.5128B262K7 tok/sFits in memory Ling-2.6-flash107B262K6 tok/sFits in memory Mistral Small 4119B262K7 tok/sFits in memory Nemotron 3 Super124B262K7 tok/sFits in memory Devstral 2 2512125B262K6 tok/sFits in memory GLM 4.6V108B131K8 tok/sFits in memory Command A111B131K7 tok/sFits in memory GLM 4.5V108B66K8 tok/sFits in memory Llama 4 Scout109B262K7 tok/sFits in memory Ling-3.0-flash128B131K6 tok/sFits in memory Hy3299B262K3 tok/sFits in memory Step 3.7 Flash201B262K3 tok/sFits in memory Trinity Large Thinking399B262K2 tok/sFits in memory Hermes 4 405B406B131K2 tok/sFits in memory Jamba Large 1.7399B131K2 tok/sFits in memory Hermes 3 405B Instruct406B131K2 tok/sFits in memory Nex-N2-Pro397B262K2 tok/sFits in memory Laguna M.1226B262K4 tok/sFits in memory Hy3 preview299B262K2 tok/sFits in memory MiMo-V2.5311B262K3 tok/sFits in memory Step 3.5 Flash199B262K2 tok/sFits in memory GLM 4.7358B131K2 tok/sFits in memory GLM 4.6357B131K2 tok/sFits in memory Llama 4 Maverick402B262K2 tok/sFits in memory Inkling Small266B262K3 tok/sFits in memory GLM 5.3 Flash321B262K3 tok/sFits in memory Nex-N2.5-Pro397B262K2 tok/sFits in memory MiniMax M3427B262K2 tok/sFits in memory MiniMax M1456B66K2 tok/sFits in memoryest MiniMax-01456B262K2 tok/sFits in memory Qwen2.5 72B Instruct72.7B33K13 tok/sFits in memory Gemma 2 27B27.2B8K35 tok/sFits in memory Nemotron 3 Ultra561B33K1 tok/sFits in memoryest Magnum v4 72B72.7B16K11 tok/sFits in memory R1 Distill Llama 70B70.6B8K11 tok/sFits in memory

Speech models it runs67 modelsTranscription and speech models that fit in memory. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.

Niagara 19m Batch.en20M26979 tok/sFits in memory Niagara 38m Batch.en38M21212 tok/sFits in memory Higgs Audio v3 STT2.7B350 tok/sFits in memory Cohere Transcribe 03 20262.1B450 tok/sFits in memory Distil Large v3.50.8B1182 tok/sFits in memory Lite Whisper Large v3 Acc1.4B576 tok/sFits in memory Owsm CTC v3.1 1B1.1B860 tok/sFits in memory Owsm CTC v3.2 ft 1B1B946 tok/sFits in memory Owsm CTC v4 1B1B806 tok/sFits in memory Hubert Large Ls960 ft0.3B2955 tok/sFits in memory MMS 1b All1B806 tok/sFits in memory Wav2vec2 Large 960h Lv60 Self0.3B2955 tok/sFits in memory Hojo ASR V15.2B156 tok/sFits in memory Granite 4.0 1b Speech2.3B351 tok/sFits in memory Granite Speech 3.3 2b3B315 tok/sFits in memory Granite Speech 4.1 2b2.3B411 tok/sFits in memory Granite Speech 4.1 2b NAR2.3B235 tok/sFits in memory STT 2.6b en2.6B310 tok/sFits in memory Phi 4 Multimodal Instruct5.6B131K169 tok/sFits in memory Voxtral Mini 3B 25074.7B172 tok/sFits in memory Voxtral Mini 4B Realtime 26024.4B183 tok/sFits in memory Canary 180m Flash0.2B4478 tok/sFits in memory Canary 1b1B806 tok/sFits in memory Canary 1b Flash0.8B1008 tok/sFits in memory Canary 1b v21B540 tok/sFits in memory Canary Qwen 2.5b2.6B310 tok/sFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6B1343 tok/sFits in memory Nemotron Speech Streaming en 0.6b0.6B1343 tok/sFits in memory Parakeet CTC 0.6b0.6B1343 tok/sFits in memory Parakeet CTC 1.1b1.1B733 tok/sFits in memory Parakeet RNNT 0.6b0.6B1343 tok/sFits in memory Parakeet RNNT 1.1b1.1B491 tok/sFits in memory Parakeet TDT CTC 110m0.1B4905 tok/sFits in memory Parakeet TDT 0.6b v20.6B1343 tok/sFits in memory Parakeet TDT 0.6b v30.6B899 tok/sFits in memory Parakeet TDT 1.1b1.1B860 tok/sFits in memory Nyra Health CrisperWhisper1.6B337 tok/sFits in memory Whisper Large v31.5B360 tok/sFits in memory Whisper Large v3 Turbo0.8B1182 tok/sFits in memory MOSS Transcribe Diarize0.9B600 tok/sFits in memory Qwen3 ASR 0.6B0.9B896 tok/sFits in memory Qwen3 ASR 0.6B HF0.8B1008 tok/sFits in memory Qwen3 ASR 1.7B2.3B351 tok/sFits in memory Qwen3 ASR 1.7B HF2B403 tok/sFits in memory Zipformer cr CTC Transducer XL 290M0.3B1861 tok/sFits in memory Zipformer Transducer XL 290M0.3B2780 tok/sFits in memory ASR Conformer Largescaleasr0.5B1970 tok/sFits in memory GLM ASR Nano 25122.3B351 tok/sFits in memory Granite Speech 5.0 470m Turboctc0.5B1891 tok/sFits in memory Granite Speech 5.0 470m Turboctc nc0.5B1079 tok/sFits in memory Fun ASR Nano 2512 HF0.8B1008 tok/sFits in memory Gemma 4 E2B it5.1B158 tok/sFits in memory Gemma 4 E4B it8B131K118 tok/sFits in memory Niagara 9m Batch.en9M59953 tok/sFits in memory Niagara 84m Batch.en84M11258 tok/sFits in memory Granite Speech 3.3 8b8.6B110 tok/sFits in memory VibeVoice ASR HF8.3B114 tok/sFits in memory ARK ASR 0.6B1.3B33K620 tok/sFits in memory ARK ASR 3B4.1B33K197 tok/sFits in memory Audio8 ASR 0.1B0.3B33K3152 tok/sFits in memory Higgs Audio v3 8b STT v28.9B106 tok/sFits in memory MOSS Transcribe Preview 2B2.4B33K336 tok/sFits in memory Gemma 4 12B it12B79 tok/sFits in memory Moonshine Streaming Medium0.3B4K2687 tok/sFits in memory Moonshine Streaming Small0.1B4K5396 tok/sFits in memory Moonshine Streaming Tiny0B4Kno speed estimateFits in memory Moonshine Tiny0B194no speed estimateFits in memory

Runs only by spilling into system memory12 modelsThese load, but part of the weights sits in ordinary system memory, which is far slower than the chip.

GLM 5.2753B—no speed estimateSpills to system RAM DeepSeek V3.2685B—no speed estimateSpills to system RAM R1 0528685B—no speed estimateSpills to system RAMest GLM 5.1754B—no speed estimateSpills to system RAM GLM 5754B—no speed estimateSpills to system RAM DeepSeek V3.2 Exp685B—no speed estimateSpills to system RAM DeepSeek V3.1 Terminus685B—no speed estimateSpills to system RAM DeepSeek V3 0324685B—no speed estimateSpills to system RAMest DeepSeek V3685B—no speed estimateSpills to system RAMest Hy4 preview780B—no speed estimateSpills to system RAM DeepSeek V4.1 Flash763B—no speed estimateSpills to system RAM DeepSeek V3.1685B—no speed estimateSpills to system RAMest

Too large for this device14 modelsThe weights do not fit even with part of them offloaded to system memory. Closest calls first.

MiMo-V2.5-Pro1T—no speed estimateToo large Ring-2.6-1T1T—no speed estimateToo large Ling-2.6-1T1T—no speed estimateToo large Kimi K2 Thinking1.1T—no speed estimateToo large Kimi K2.51.1T—no speed estimateToo large Kimi K2.61.1T—no speed estimateToo large Inkling952B—no speed estimateToo large Kimi K2 09051T—no speed estimateToo large Kimi K2 07111T—no speed estimateToo large Kimi K2.7 Code1.1T—no speed estimateToo large DeepSeek V4 Pro1.6T—no speed estimateToo large LongCat 2.01.8T—no speed estimateToo large Qwen3.8 2.4T A95B2.4T—no speed estimateToo large Kimi K32.8T—no speed estimateToo large

224 models have a verdict on this device: 198 fit in memory, 12 spill into system memory and 14 are too large. Every one of them is on this page.

03

Other M3 chips

The same generation in other memory sizes. Memory is what decides which models fit.

Apple M3 (10-core GPU)24 GB · less memory Apple M3 Pro (18-core GPU)36 GB · less memory Apple M3 Max (40-core GPU)128 GB · less memory

Something wrong on this page? Tell us