AI Index

Open-weight models that run on Apple M4 · 32 GB

  • Unified memory 32 GB
  • Bandwidth 120 GB/s
  • Usable 21.3 GB

Specs: Memory · MacBook Air (13-inch, M4, 2025) - Tech Specs - Apple Support · Bandwidth · MacBook Air (13-inch, M4, 2025) - Tech Specs - Apple Support

FitModelParameters / activePrecisionMemoryEst. speedLongest contextRuns as
Runs wellQwen3.5-35B-A3B Alibaba Qwen36B / 3BIQ4_XS18.8 / 21.3 GB34–51 tok/s136KUnified memory
Runs wellQwen3.6-35B-A3B Alibaba Qwen36B / 3BIQ4_XS18.8 / 21.3 GB34–51 tok/s136KUnified memory
Runs wellQwen3-Omni-30B-A3B-Captioner Alibaba Qwen31.7BQ4_018.2 / 21.3 GB18–27 tok/s41KUnified memory
Runs wellGLM-4.7-Flash Zhipu AI31.2BQ4_K_M19.0 / 21.3 GB17–25 tok/s54KUnified memory
Runs wellQwen3-VL-30B-A3B-Instruct Alibaba Qwen31.1BQ4_K_Mmaker’s file18.9 / 21.3 GB19–29 tok/s34KUnified memory
Runs wellQwen3-VL-30B-A3B-Thinking Alibaba Qwen31.1BQ4_K_Mmaker’s file18.9 / 21.3 GB19–29 tok/s34KUnified memory
Runs wellQwen3-30B-A3B-Instruct-2507 Alibaba Qwen30.5B / 3.3BQ4_K_M18.9 / 21.3 GB21–32 tok/s34KUnified memory
Runs wellQwen3-30B-A3B-Thinking-2507 Alibaba Qwen30.5B / 3.3BQ4_K_M18.9 / 21.3 GB21–32 tok/s34KUnified memory
Runs wellQwen3-30B-A3B Alibaba Qwen30.5B / 3.3BQ4_K_Mmaker’s file18.9 / 21.3 GB21–32 tok/s34KUnified memory
Runs wellQwen3-Coder-30B-A3B-Instruct Alibaba Qwen30.5B / 3.3BQ4_K_M18.9 / 21.3 GB21–32 tok/s34KUnified memory
Runs wellHy-Memory-30B-A3B Tencent30.1B / 3BQ4_K_Mmaker’s file18.6 / 21.3 GB23–34 tok/s37KUnified memory
Runs wellGemma 4 26B A4B IT Google25.8B / 3.8BQ5_K_M18.5 / 21.3 GB18–28 tok/s81KUnified memory
Runs wellERNIE-4.5-21B-A3B-Thinking Baidu21.8B / 3BQ6_K18.0 / 21.3 GB20–31 tok/s69KUnified memory
Runs wellGPT OSS 20B OpenAI20.9B / 3.6BMXFP4 (native)maker’s file13.6 / 21.3 GB23–35 tok/s128KUnified memory
Runs wellpplx-embed-v2-context-9b Perplexity8.4BIQ4_XS4.9 / 21.3 GB13–19 tok/s256KUnified memory
Runs wellGemma 4 E4B Google8BQ4_04.8 / 21.3 GB13–19 tok/s125KUnified memory
Runs wellLing-3.0-tiny Ant Group7.9B / 1.3BBF16 (native)maker’s file15.6 / 21.3 GB21–32 tok/s128KUnified memory
Runs wellQwen2.5-7B-Instruct-1M Alibaba Qwen7.6BQ4_04.9 / 21.3 GB13–19 tok/s308KUnified memory
Runs wellQwen2.5-7B-Instruct Alibaba Qwen7.6BIQ4_XS4.7 / 21.3 GB13–20 tok/s32KUnified memory
Runs wellDeepSeek-R1-Distill-Qwen-7B DeepSeek7.6BQ4_04.9 / 21.3 GB13–19 tok/s32KUnified memory
Runs wellGemma 4 E2B Google5.1BQ6_K4.5 / 21.3 GB14–21 tok/s125KUnified memory
Runs wellWeVisOne-VL-4B Tencent4.8BQ5_K_M4.8 / 21.3 GB13–19 tok/s125KUnified memory
Runs wellQianfan-OCR Baidu4.7BQ5_K_M4.8 / 21.3 GB13–20 tok/s31KUnified memory
Runs wellQwen3.5-4B Alibaba Qwen4.7BQ6_K4.3 / 21.3 GB15–22 tok/s256KUnified memory
Runs wellQwen3-VL-4B-Instruct Alibaba Qwen4.4BQ5_K_M4.6 / 21.3 GB14–21 tok/s127KUnified memory
Runs wellQwen3-4B-Instruct-2507 Alibaba Qwen4BQ6_K4.7 / 21.3 GB13–20 tok/s126KUnified memory
Runs wellQwen3-4B Alibaba Qwen4BQ6_Kmaker’s file4.7 / 21.3 GB13–20 tok/s128KUnified memory
Runs wellDeepSeek-OCR DeepSeek3.3BBF16 (native)maker’s file7.2 / 21.3 GB22–32 tok/s8KUnified memory
Runs wellGranite-4.0-H-Micro IBM3.2BQ8_0maker’s file3.7 / 21.3 GB17–26 tok/s128KUnified memory
Runs wellQwen3-1.7B Alibaba Qwen2BMLX 8bitmaker’s file3.0 / 21.3 GB22–33 tok/s40KUnified memory
Runs wellDeepSeek-R1-Distill-Qwen-1.5B DeepSeek1.8BBF16 (native)maker’s file4.0 / 21.3 GB16–24 tok/s125KUnified memory
Runs wellGLM-OCR Zhipu AI1.3BBF16 (native)maker’s file3.5 / 21.3 GB19–28 tok/s128KUnified memory
Runs wellQwen3-0.6B Alibaba Qwen0.8BBF16 (native)maker’s file2.8 / 21.3 GB25–37 tok/s40KUnified memory
Runs wellEmbeddingGemma 2 Google0.7BBF16 (native)maker’s file2.0 / 21.3 GB38–57 tok/s256KUnified memory
Runs wellQwen3-Embedding-0.6B Alibaba Qwen0.6BBF16 (native)maker’s file2.5 / 21.3 GB28–42 tok/s32KUnified memory
Runs wellbge-reranker-v2-m3 BAAI0.6BFP32 (native)maker’s file3.4 / 21.3 GB20–29 tok/s8KUnified memory
Runs wellbge-m3-unsupervised BAAI0.6BFP32 (native)maker’s file3.4 / 21.3 GB20–29 tok/s8KUnified memory
Runs wellbge-large-en-v1.5 BAAI0.3BFP32 (native)maker’s file1.8 / 21.3 GB43–65 tok/s512Unified memory
Runs wellbge-reranker-base BAAI0.3BFP32 (native)maker’s file1.6 / 21.3 GB53–80 tok/s514Unified memory
Runs wellbge-base-en-v1.5 BAAI0.1BFP32 (native)maker’s file0.9 / 21.3 GB131–197 tok/s512Unified memory
Runs wellbge-small-en-v1.5 BAAI0BFP32 (native)maker’s file0.6 / 21.3 GB420–630 tok/s512Unified memory
Runs, slowlyQwen2.5-VL-32B-Instruct Alibaba Qwen33.5BQ4_020.4 / 21.3 GB3–4 tok/s12KUnified memory
Runs, slowlyQwen3-VL-32B-Instruct Alibaba Qwen33.4BQ4_K_Mmaker’s file21.3 / 21.3 GB3–4 tok/s8KUnified memory
Runs, slowlyQwen3-VL-32B-Thinking Alibaba Qwen33.4BQ4_K_Mmaker’s file21.3 / 21.3 GB3–4 tok/s8KUnified memory
Runs, slowlyQwen2.5-32B-Instruct Alibaba Qwen32.8BIQ4_XS19.0 / 21.3 GB3–5 tok/s17KUnified memory
Runs, slowlyQwen2.5-Coder-32B-Instruct Alibaba Qwen32.8BIQ4_XS19.0 / 21.3 GB3–5 tok/s17KUnified memory
Runs, slowlyQwQ-32B Alibaba Qwen32.8BIQ4_XS19.0 / 21.3 GB3–5 tok/s17KUnified memory
Runs, slowlyDeepSeek-R1-Distill-Qwen-32B DeepSeek32.8BIQ4_XS19.0 / 21.3 GB3–5 tok/s17KUnified memory
Runs, slowlyQwen3-32B Alibaba Qwen32.8BMLX 4bitmaker’s file19.0 / 21.3 GB3–5 tok/s17KUnified memory
Runs, slowlyGemma 4 31B IT Google31.3BIQ4_XS18.3 / 21.3 GB3–5 tok/s27KUnified memory
Runs, slowlyGranite Switch 4.1 30B IBM30BIQ4_XS17.6 / 21.3 GB3–5 tok/s22KUnified memory
Runs, slowlyQwen3.5-27B Alibaba Qwen27.8BQ5_019.2 / 21.3 GB3–5 tok/s42KUnified memory
Runs, slowlyQwen3.6-27B Alibaba Qwen27.8BQ5_019.2 / 21.3 GB3–5 tok/s42KUnified memory
Runs, slowlyQwen3.8-27B Alibaba Qwen27.8BQ5_019.2 / 21.3 GB3–5 tok/s42KUnified memory
Runs, slowlyMagistral Small 1.2 Mistral AI24BQ5_K_Mmaker’s file17.6 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyDevstral Small 1.0 Mistral AI23.6BQ5_K_M17.6 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyDevstral Small 1.1 Mistral AI23.6BQ5_K_M17.6 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyMagistral Small 1.0 Mistral AI23.6BQ5_K_M17.6 / 21.3 GB3–5 tok/s31KUnified memory
Runs, slowlyMagistral Small 1.1 Mistral AI23.6BQ5_K_Mmaker’s file17.6 / 21.3 GB3–5 tok/s31KUnified memory
Runs, slowlyQwen2.5-14B-Instruct-1M Alibaba Qwen14.8BQ8_016.8 / 21.3 GB3–5 tok/s31KUnified memory
Runs, slowlyQwen2.5-14B-Instruct Alibaba Qwen14.8BQ8_0maker’s file16.9 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyDeepSeek-R1-Distill-Qwen-14B DeepSeek14.8BQ8_016.8 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyQwen3-14B Alibaba Qwen14.8BMLX 8bitmaker’s file16.1 / 21.3 GB4–5 tok/s128KUnified memory
Runs, slowlyGemma 3 12B IT Google12.2BQ8_015.5 / 21.3 GB4–6 tok/s24KUnified memory
Runs, slowlyQwen2.5-Omni-7B Alibaba Qwen10.7BQ8_011.6 / 21.3 GB5–8 tok/s32KUnified memory
Runs, slowlyGLM-4.6V-Flash Zhipu AI10.3BQ8_011.0 / 21.3 GB5–8 tok/s125KUnified memory
Runs, slowlyQwen3.5-9B Alibaba Qwen9.7BBF16 (native)maker’s file19.1 / 21.3 GB3–5 tok/s78KUnified memory
Runs, slowlyQianfan-VL-8B Baidu8.8BBF16 (native)maker’s file18.2 / 21.3 GB3–5 tok/s32KUnified memory
Runs, slowlyGranite Switch 4.1 8B IBM8.8BBF16 (native)maker’s file18.5 / 21.3 GB3–5 tok/s25KUnified memory
Runs, slowlyQwen3-VL-8B-Instruct Alibaba Qwen8.8BBF16 (native)maker’s file18.3 / 21.3 GB3–5 tok/s29KUnified memory
Runs, slowlyQwen3-VL-8B-Thinking Alibaba Qwen8.8BBF16 (native)maker’s file18.3 / 21.3 GB3–5 tok/s29KUnified memory
Runs, slowlyQwen2.5-VL-7B-Instruct Alibaba Qwen8.3BBF16 (native)maker’s file16.7 / 21.3 GB4–5 tok/s94KUnified memory
Runs, slowlyQwen3-8B Alibaba Qwen8.2BBF16 (native)maker’s file17.1 / 21.3 GB3–5 tok/s38KUnified memory
Runs, slowlyDeepSeek-R1-Distill-Llama-8B DeepSeek8BBF16 (native)maker’s file16.7 / 21.3 GB4–5 tok/s45KUnified memory
Runs, slowlyMistral-7B-Instruct-v0.1 Mistral AI7.2BBF16 (native)maker’s file15.2 / 21.3 GB4–6 tok/s32KUnified memory

Estimates from the makers’ Hugging Face repositories (as of 2026-11-06); filtered views aren’t indexed. How it’s estimated