Hardware

Apple M3 Pro

Apple M3 Pro has 18 / 36 GB Unified and 150 GB/s. See which open models fit, at which quant, and roughly how fast. Figures below use the default memory (18 GB). Updated 2026-10-24.

Memory18 / 36 GB Unified

Bandwidth150 GB/s

Kindmac

Year2023

150 GB/s. 18GB or 36GB. Source.

RSS for new models on this device

Models that fit

Model	Verdict	Quant	Context	tok/s
gpt oss 20b 20.9B	Tight fit	Q8_0	8K	22–30
GLM 4.7 Flash REAP 23B A3B 23.0B	Tight fit	IQ4_XS	8K	40–54
Qwen3.5 35B A3B 34.7B	Tight fit	UD-IQ3_XXS	4K	68–93
gemma 4 26B A4B it 25.2B	Tight fit	Q3_K_M	4K	27–37
Qwen3.6 14B A3B FableVibes 13.8B	Tight fit	Q6_K	8K	31–43
LFM2.5 24B A2B 24.0B	Tight fit	UD-Q3_K_XL	8K	52–71
GLM 4.7 Flash 29.9B	Tight fit	UD-IQ3_XXS	4K	28–38
Qwen3.6 35B A3B 34.7B	Tight fit	UD-Q2_K_XL	8K	67–91
Tiel Coder 35B A3B 34.7B	Tight fit	UD-Q2_K_XL	8K	67–91
bottlecapai ThinkingCap gemma 4 26B A4B it 25.2B	Tight fit	Q3_K_S	8K	21–28
Qwen3 Coder 30B A3B Instruct 30.5B	Tight fit	UD-Q2_K_XL	8K	42–57
Qwen3 30B A3B Instruct 2507 30.5B	Tight fit	UD-Q2_K_XL	8K	42–57
Qwen3 VL 30B A3B Instruct 30.5B	Tight fit	UD-Q2_K_XL	8K	42–57
Qwen3 30B A3B Thinking 2507 30.5B	Tight fit	UD-Q2_K_XL	8K	42–57
Kwaipilot KAT Coder V2.5 Dev 34.7B	Runs	IQ2_S	8K	38–52
Qwen2.5 7B Instruct 7.6B	Runs	Q8_0	8K	9.6–13
DeepSeek R1 Distill Qwen 7B 7.6B	Runs	Q8_0	8K	9.6–13
Qwen2.5 Coder 7B Instruct 128K 7.6B	Runs	Q8_0	8K	9.6–13
gemma 4 E4B it qat 7.5B	Runs great	UD-Q4_K_XL	8K	17–23
LFM2.5 8B A1B 8.5B	Runs	UD-Q8_K_XL	8K	55–75
Sharp Spark X2.5 4B 4.1B	Runs great	Q6_K	8K	17–23
granite 4.2 3b 3.7B	Runs	BF16	8K	10–14
Ministral 3 3B Instruct 2512 3.4B	Runs great	BF16	8K	11–15
Ministral 3 3B Reasoning 2512 3.4B	Runs great	BF16	8K	11–15
granite 4.0 h micro 3.2B	Runs great	BF16	8K	13–17
Llama 3.2 3B Instruct 3.2B	Runs great	BF16	8K	11–15
Qwen2.5 3B Instruct 3.1B	Runs great	F16	8K	13–17
Qwen2.5 VL 3B Instruct 3.1B	Runs great	BF16	8K	13–17
SmolLM3 3B 3.1B	Runs great	BF16	8K	12–17
LFM2.5 2.6B 2.7B	Runs great	BF16	8K	14–19
LFM2.5 VL 3B 2.7B	Runs great	BF16	8K	14–19
Qwen3.8 2B Distill 1.9B	Runs great	BF16	8K	21–28
Qwen3.5 2B 1.9B	Runs great	BF16	8K	21–29
Hy MT2 1.8B 1.8B	Runs great	Q8_0	8K	34–46
DeepSeek R1 Distill Qwen 1.5B 1.8B	Runs great	BF16	8K	22–30
Qwen3 1.7B 1.7B	Runs great	BF16	8K	19–26
Qwen3 VL 2B Instruct 1.7B	Runs great	BF16	8K	19–26
Qwen3 VL 2B Thinking 1.7B	Runs great	BF16	8K	19–26
Qwen2.5 1.5B Instruct 1.5B	Runs great	F16	8K	25–34
Qwen2.5 Coder 1.5B Instruct 1.5B	Runs great	F16	8K	25–34

105 more models in the catalog do not fit this default profile, or ranked below the first 40 that do.

Example: Tight fit fully in unified memory on your Apple M3 Pro at Q8_0, 8K context, about 22–30 tok/s. Weights 11.3 GB plus 0.19 GB of KV cache — 11.9 GB of 13.0 GB usable. Up to 32K context still fits. Weights shown in the verdict use GiB (the same 1024³ counting as GPU memory).