Hardware / Apple /Apple M2 (8-core GPU, 8GB unified)
Apple · Apple Silicon
Apple M2 (8-core GPU, 8GB unified)
for example MacBook Air (2022), MacBook Pro 13-inch (2022), Mac mini (2023). Memory is a build option on these, so not every one of them is sold with 8 GB.
Runs 69 models fully in memory, 18 more with CPU offload.
Memory
8 GB
Bandwidth
100 GB/s
Runs
69
01
Specifications
Memory
8 GB
Decides which models fit at all.
Memory bandwidth
100 GB/s
How fast it can read the model. Sets the speed of the reply.
Compute
5.6 TFLOPS
Trillions of half-precision sums a second. Sets how fast it reads a long prompt, not how fast it replies.
Power draw
15 W
At full load, for sizing a power supply.
Released
Jun 24, 2022
Architecture
M2
02
What is quantisation? →What it runs
Llama 3.2 1B Instruct1.2B33K96 tok/sFits in memoryLlama 3.2 3B Instruct3.2B66K36 tok/sFits in memoryGranite 4.0 Micro3.2B33K36 tok/sFits in memoryGemma 3 4B4.3B33K27 tok/sFits in memoryHy-MT2-1.8B2B8K49 tok/sFits in memoryMinistral 3 3B 25123.8B16K30 tok/sFits in memoryNemotron 3.5 Content Safety4.3B8K27 tok/sFits in memory
Speech models it runs62 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.en20M5773 tok/sFits in memoryNiagara 38m Batch.en38M1734 tok/sFits in memoryDistil Large v3.50.8B123 tok/sFits in memoryHubert Large Ls960 ft0.3B308 tok/sFits in memoryWav2vec2 Large 960h Lv60 Self0.3B206 tok/sFits in memoryCanary 180m Flash0.2B642 tok/sFits in memoryCanary 1b Flash0.8B123 tok/sFits in memoryNemotron 3.5 ASR Streaming 0.6b0.6B192 tok/sFits in memoryNemotron Speech Streaming en 0.6b0.6B192 tok/sFits in memoryParakeet CTC 0.6b0.6B192 tok/sFits in memoryParakeet RNNT 0.6b0.6B192 tok/sFits in memoryParakeet TDT CTC 110m0.1B1050 tok/sFits in memoryParakeet TDT 0.6b v20.6B192 tok/sFits in memoryParakeet TDT 0.6b v30.6B192 tok/sFits in memoryWhisper Large v3 Turbo0.8B123 tok/sFits in memoryQwen3 ASR 0.6B0.9B128 tok/sFits in memoryZipformer cr CTC Transducer XL 290M0.3B398 tok/sFits in memoryZipformer Transducer XL 290M0.3B398 tok/sFits in memoryASR Conformer Largescaleasr0.5B137 tok/sFits in memoryGranite Speech 5.0 470m Turboctc0.5B231 tok/sFits in memoryGranite Speech 5.0 470m Turboctc nc0.5B132 tok/sFits in memoryNiagara 9m Batch.en9M10935 tok/sFits in memoryNiagara 84m Batch.en84M784 tok/sFits in memoryOwsm CTC v3.2 ft 1B1B116 tok/sFits in memoryOwsm CTC v4 1B1B116 tok/sFits in memoryMMS 1b All1B116 tok/sFits in memoryCanary 1b1B116 tok/sFits in memoryCanary 1b v21B116 tok/sFits in memoryAudio8 ASR 0.1B0.3B33K385 tok/sFits in memoryMOSS Transcribe Diarize0.9B128 tok/sFits in memoryQwen3 ASR 0.6B HF0.8B144 tok/sFits in memoryFun ASR Nano 2512 HF0.8B123 tok/sFits in memoryOwsm CTC v3.1 1B1.1B105 tok/sFits in memoryParakeet CTC 1.1b1.1B105 tok/sFits in memoryParakeet RNNT 1.1b1.1B105 tok/sFits in memoryParakeet TDT 1.1b1.1B105 tok/sFits in memoryLite Whisper Large v3 Acc1.4B83 tok/sFits in memoryARK ASR 0.6B1.3B33K89 tok/sFits in memoryWhisper Large v31.5B77 tok/sFits in memoryNyra Health CrisperWhisper1.6B72 tok/sFits in memoryMoonshine Streaming Medium0.3B4K220 tok/sFits in memoryMoonshine Streaming Small0.1B4K1155 tok/sFits in memoryCohere Transcribe 03 20262.1B55 tok/sFits in memoryQwen3 ASR 1.7B2.3B50 tok/sFits in memorySTT 2.6b en2.6B44 tok/sFits in memoryCanary Qwen 2.5b2.6B44 tok/sFits in memoryGranite 4.0 1b Speech2.3B50 tok/sFits in memoryGranite Speech 4.1 2b2.3B50 tok/sFits in memoryGranite Speech 4.1 2b NAR2.3B50 tok/sFits in memoryGLM ASR Nano 25122.3B50 tok/sFits in memoryQwen3 ASR 1.7B HF2B58 tok/sFits in memoryGranite Speech 3.3 2b3B39 tok/sFits in memoryHiggs Audio v3 STT2.7B43 tok/sFits in memoryARK ASR 3B4.1B33K28 tok/sFits in memoryHojo ASR V15.2B22 tok/sFits in memoryGemma 4 E2B it5.1B23 tok/sFits in memoryVoxtral Mini 4B Realtime 26024.4B26 tok/sFits in memoryMOSS Transcribe Preview 2B2.4B4K48 tok/sFits in memoryVoxtral Mini 3B 25074.7B25 tok/sFits in memoryPhi 4 Multimodal Instruct5.6B8K21 tok/sFits in memoryMoonshine Streaming Tiny0B4Kno speed estimateFits in memoryMoonshine Tiny0B194no speed estimateFits in memory
Runs only by spilling into system memory18 modelsThese load, but part of the weights sits in ordinary system memory, which is far slower than the chip.
Granite 4.1 8B8.8B—no speed estimateSpills to system RAMReka Edge7.1B—no speed estimateSpills to system RAMestQwen3.5-9B9.7B—no speed estimateSpills to system RAMestMinistral 3 8B 25128.9B—no speed estimateSpills to system RAMQwen3 VL 8B Thinking8.8B—no speed estimateSpills to system RAMUI-TARS 7B8.3B—no speed estimateSpills to system RAMGemma 3n 4B7.8B—no speed estimateSpills to system RAMLlama 3 8B Lunaris8B—no speed estimateSpills to system RAMQwen3 VL 8B Instruct8.8B—no speed estimateSpills to system RAMQwen3 8B8.2B—no speed estimateSpills to system RAMQwen2.5 7B Instruct7.6B—no speed estimateSpills to system RAMestLlama 3.1 8B Instruct8B—no speed estimateSpills to system RAMestHiggs Audio v3 8b STT v28.9B—no speed estimateSpills to system RAMGranite Speech 3.3 8b8.6B—no speed estimateSpills to system RAMVibeVoice ASR HF8.3B—no speed estimateSpills to system RAMHy-MT2-7B8B—no speed estimateSpills to system RAMGranite 4.2 8B8.8B—no speed estimateSpills to system RAMGemma 4 E4B it8B—no speed estimateSpills to system RAM
Too large for this device137 modelsThe weights do not fit even with part of them offloaded to system memory. Closest calls first.
Gemma 3 12B12.2B—no speed estimateToo largeestRocinante 12B12.2B—no speed estimateToo largeestLlama Guard 4 12B12B—no speed estimateToo largeUnslopNemo 12B12.2B—no speed estimateToo largeMistral Nemo12.2B—no speed estimateToo largeMinistral 3 14B 251213.9B—no speed estimateToo largeQwen3 14B14.8B—no speed estimateToo largegpt-oss-safeguard-20b21.5B—no speed estimateToo largeGemma 4 12B it12B—no speed estimateToo largeCydonia 24B V4.123.6B—no speed estimateToo largeUncensored24B—no speed estimateToo largeReka Flash 320.9B—no speed estimateToo largeReMM SLERP 13B13B—no speed estimateToo largeMythoMax 13B13B—no speed estimateToo largeMistral Small 323.6B—no speed estimateToo largeQwen3.5-27B27.8B—no speed estimateToo largeMistral Small 3.1 24B24B—no speed estimateToo largeQwen3 30B A3B Instruct 250730.5B—no speed estimateToo largeQwen3 VL 30B A3B Thinking31.1B—no speed estimateToo largeNemotron 3.5 Lightning31.6B—no speed estimateToo largeTernary Bonsai 2 27B27B—no speed estimateToo largeGemma 3 27B27.4B—no speed estimateToo largeQwen3.8 27B27.8B—no speed estimateToo largeMuse Glimmer 30B29.8B—no speed estimateToo largeQwen3.5-35B-A3B36B—no speed estimateToo largeMistral Small 3.2 24B24B—no speed estimateToo largeVoxtral Small 24B 250724.3B—no speed estimateToo largeSkyfall 36B V236.9B—no speed estimateToo largeGemma 4 26B A4B25.8B—no speed estimateToo largePhi 414.7B—no speed estimateToo largeGemma 2 27B27.2B—no speed estimateToo largeQwen3.6 27B27.8B—no speed estimateToo largeGLM 4.7 Flash31.2B—no speed estimateToo largeHy-MT2-30B-A3B30.1B—no speed estimateToo largeQwen3 30B A3B Thinking 250730.5B—no speed estimateToo largeQwen3 Coder 30B A3B Instruct30.5B—no speed estimateToo largeQwen3 VL 30B A3B Instruct31.1B—no speed estimateToo largeNemotron 3 Nano 30B A3B31.6B—no speed estimateToo largeLaguna XS 2.133.4B—no speed estimateToo largeOlmo 3 32B Think32.2B—no speed estimateToo largeGemma 4 31B31.3B—no speed estimateToo largeQwen2.5 Coder 32B Instruct32.8B—no speed estimateToo largeQwen3 32B32.8B—no speed estimateToo largeQwen3 VL 32B Instruct33.4B—no speed estimateToo largeNex-N2-Mini35.1B—no speed estimateToo largeNex-N2.5-Mini35.1B—no speed estimateToo largeQwen3.6 35B A3B36B—no speed estimateToo largeLlama 3.1 70B Instruct70.6B—no speed estimateToo largeHermes 3 70B Instruct70.6B—no speed estimateToo largeQwen2.5 72B Instruct72.7B—no speed estimateToo largeQwen3 Coder Next79.7B—no speed estimateToo largeHunyuan A13B Instruct80.4B—no speed estimateToo largeQwen3 Next 80B A3B Thinking81.3B—no speed estimateToo largeQwen3 Next 80B A3B Instruct81.3B—no speed estimateToo largeLing-2.6-flash107B—no speed estimateToo largeCommand A111B—no speed estimateToo largegpt-oss-120b120B—no speed estimateToo largeNemotron 3 Super124B—no speed estimateToo largeLlama 3.3 70B Instruct70.6B—no speed estimateToo largeHermes 4 70B70.6B—no speed estimateToo largeLlama 3.3 Euryale 70B70.6B—no speed estimateToo largeLlama 3.1 Euryale 70B v2.270.6B—no speed estimateToo largeR1 Distill Llama 70B70.6B—no speed estimateToo largeQwen3.5-122B-A10B125B—no speed estimateToo largeDevstral 2 2512125B—no speed estimateToo largeGLM 4.6V108B—no speed estimateToo largeGLM 4.5V108B—no speed estimateToo largeLlama 4 Scout109B—no speed estimateToo largeMagnum v4 72B72.7B—no speed estimateToo largeMistral Medium 3.5128B—no speed estimateToo largeQwen2.5 VL 72B Instruct73.4B—no speed estimateToo largeLaguna S 2.1118B—no speed estimateToo largeMistral Small 4119B—no speed estimateToo largeLing-3.0-flash128B—no speed estimateToo largeMixtral 8x22B Instruct141B—no speed estimateToo largeGLM 4.5 Air111B—no speed estimateToo largeStep 3.7 Flash201B—no speed estimateToo largeMiniMax M2229B—no speed estimateToo largeStep 3.5 Flash199B—no speed estimateToo largeMiniMax M2.5229B—no speed estimateToo largeDeepSeek V4 Flash291B—no speed estimateToo largeDeepSeek V4 Flash Vision Exp305B—no speed estimateToo largeHy3299B—no speed estimateToo largeGLM 4.5358B—no speed estimateToo largeLaguna M.1226B—no speed estimateToo largeNex-N2.5-Pro397B—no speed estimateToo largeMiniMax M2.7229B—no speed estimateToo largeMiniMax M2.1229B—no speed estimateToo largeQwen3 235B A22B Thinking 2507235B—no speed estimateToo largeQwen3 235B A22B Instruct 2507235B—no speed estimateToo largeQwen3 VL 235B A22B Thinking236B—no speed estimateToo largeQwen3 VL 235B A22B Instruct236B—no speed estimateToo largeGLM 4.6357B—no speed estimateToo largeGLM 4.7358B—no speed estimateToo largeMiniMax M3427B—no speed estimateToo largeMiniMax-01456B—no speed estimateToo largeInkling Small266B—no speed estimateToo largeJamba Large 1.7399B—no speed estimateToo largeLlama 4 Maverick402B—no speed estimateToo largeERNIE 4.5 VL 424B A47B424B—no speed estimateToo largeHy3 preview299B—no speed estimateToo largeMiniMax M1456B—no speed estimateToo largeMiMo-V2.5311B—no speed estimateToo largeGLM 5.3 Flash321B—no speed estimateToo largeQwen3 Coder 480B A35B480B—no speed estimateToo largeDeepSeek V3 0324685B—no speed estimateToo largeDeepSeek V3.2685B—no speed estimateToo largeNex-N2-Pro397B—no speed estimateToo largeTrinity Large Thinking399B—no speed estimateToo largeQwen3.5 397B A17B403B—no speed estimateToo largeHermes 4 405B406B—no speed estimateToo largeHermes 3 405B Instruct406B—no speed estimateToo largeGLM 5.1754B—no speed estimateToo largeHy4 preview780B—no speed estimateToo largeR1 0528685B—no speed estimateToo largeDeepSeek V3685B—no speed estimateToo largeGLM 5754B—no speed estimateToo largeGLM 5.2753B—no speed estimateToo largeNemotron 3 Ultra561B—no speed estimateToo largeMiMo-V2.5-Pro1T—no speed estimateToo largeKimi K2 07111T—no speed estimateToo largeLing-2.6-1T1T—no speed estimateToo largeDeepSeek V3.1 Terminus685B—no speed estimateToo largeDeepSeek V3.1685B—no speed estimateToo largeDeepSeek V3.2 Exp685B—no speed estimateToo largeRing-2.6-1T1T—no speed estimateToo largeKimi K2 Thinking1.1T—no speed estimateToo largeKimi K2.7 Code1.1T—no speed estimateToo largeKimi K2.61.1T—no speed estimateToo largeKimi K2.51.1T—no speed estimateToo largeDeepSeek V4.1 Flash763B—no speed estimateToo largeInkling952B—no speed estimateToo largeKimi K2 09051T—no speed estimateToo largeDeepSeek V4 Pro1.6T—no speed estimateToo largeLongCat 2.01.8T—no speed estimateToo largeQwen3.8 2.4T A95B2.4T—no speed estimateToo largeKimi K32.8T—no speed estimateToo large
224 models have a verdict on this device: 69 fit in memory, 18 spill into system memory and 137 are too large. Every one of them is on this page.
03