What LLMs can Mac Studio M2 Ultra 192 GB run?
192 GB unified, 800 GB/s memory bandwidth, about $6,599. Every open model in our catalog, rated by the engine for chat and for agentic coding.
Runs well (42)
Fast enough and with memory to spare for everyday chat.
| Model | Chat | Agentic coding | Quant | Chat speed |
|---|---|---|---|---|
| gpt-oss-120bOpenAIOpen source | Excellent | Excellent | MXFP4 | 85–120 tok/s |
| Qwen3.6 35B-A3BAlibaba QwenOpen source | Excellent | Excellent | 6-bit | 100–130 tok/s |
| Qwen3-Next 80B-A3B InstructAlibaba QwenOpen source | Excellent | Comfortable | 6-bit | 100–130 tok/s |
| Gemma 4 26B-A4BGoogleOpen source | Excellent | Comfortable | 6-bit | 80–110 tok/s |
| Qwen3.5 35B-A3BAlibaba QwenOpen source | Excellent | Comfortable | 6-bit | 100–130 tok/s |
| LFM2.5 24B-A2BLiquid AI | Excellent | Comfortable | Q4_K_M | 190–260 tok/s |
| Qwen3.5 122B-A10BAlibaba QwenOpen source | Excellent | Comfortable | 5-bit | 35–50 tok/s |
| gpt-oss-20bOpenAIOpen source | Excellent | Comfortable | MXFP4 | 120–160 tok/s |
| GLM-4.7-Flash (30B-A3B)Z.aiOpen source | Excellent | Comfortable | 6-bit | 95–130 tok/s |
| Mistral Small 4 (119B-A6.5B)Mistral AIOpen source | Excellent | Comfortable | 6-bit | 45–60 tok/s |
| Qwen3 30B-A3B Instruct 2507Alibaba QwenOpen source | Excellent | Comfortable | 6-bit | 85–110 tok/s |
| Kolibri-1 (78B-A3.5B)Aleph AlphaOpen source | Excellent | Acceptable | FP8 | 70–95 tok/s |
| Nemotron 3.5 Super VL 120B-A12BNVIDIA | Excellent | Comfortable | MXFP4 | 35–45 tok/s |
| Qwen3.8 27BAlibaba QwenOpen source | Excellent | Acceptable | 4-bit | 20–30 tok/s |
| GLM-4.5-Air (106B-A12B)Z.aiOpen source | Excellent | Acceptable | 5-bit | 30–40 tok/s |
| Qwen3-Coder 30B-A3B InstructAlibaba QwenOpen source | Excellent | Comfortable | 6-bit | 85–110 tok/s |
| MiniCPM5 2BOpenBMBOpen source | Excellent | Acceptable | Q8_0 | 120–160 tok/s |
| Qwen3.6 27BAlibaba QwenOpen source | Excellent | Acceptable | 4-bit | 20–30 tok/s |
| Qwen3.5 9BAlibaba QwenOpen source | Excellent | Acceptable | 6-bit | 40–55 tok/s |
| Gemma 4 12BGoogleOpen source | Excellent | Acceptable | 5-bit | 40–55 tok/s |
| Qwen3.5 4BAlibaba QwenOpen source | Excellent | Acceptable | 6-bit | 80–110 tok/s |
| Nemotron 3.5 Bolt 74B-A9BNVIDIA | Excellent | Acceptable | 4-bit | 50–65 tok/s |
| Qwen3.5 27BAlibaba QwenOpen source | Excellent | Acceptable | 4-bit | 20–30 tok/s |
| Gemma 4 31BGoogleOpen source | Excellent | Acceptable | 4-bit | 18–25 tok/s |
| GLM-5.3-Flash (320B-A18B)Z.aiOpen source | Excellent | Acceptable | 3-bit | 30–40 tok/s |
| GLM-4.6V-Flash (10B)Z.aiOpen source | Excellent | Acceptable | 6-bit | 40–50 tok/s |
| Qwen3 8BAlibaba QwenOpen source | Excellent | Acceptable | 6-bit | 45–60 tok/s |
| Qwen3 14BAlibaba QwenOpen source | Excellent | Acceptable | 5-bit | 30–40 tok/s |
| DeepSeek-R1-0528-Qwen3-8BDeepSeekOpen source | Excellent | Acceptable | 6-bit | 45–60 tok/s |
| Gemma 4 E4BGoogleOpen source | Excellent | Borderline | 6-bit | 50–65 tok/s |
| Phi-4 (14B)MicrosoftOpen source | Excellent | Acceptable | 5-bit | 30–40 tok/s |
| Llama 4 Scout (109B-A17B)MetaOpen weights (restricted) | Excellent | Acceptable | 4-bit | 25–35 tok/s |
| Muse Glimmer (30B)MetaOpen source | Excellent | Acceptable | Q4_K_M | 20–25 tok/s |
| Llama 3.2 3B InstructMetaOpen weights (restricted) | Excellent | Technically Runs | 6-bit | 110–150 tok/s |
| Llama 3.1 8B InstructMetaOpen weights (restricted) | Excellent | Borderline | 6-bit | 45–65 tok/s |
| Mistral Small 3.2 (24B)Mistral AIOpen source | Excellent | Acceptable | 4-bit | 25–30 tok/s |
| Gemma 3 27BGoogleOpen weights (restricted) | Excellent | Acceptable | 4-bit | 20–30 tok/s |
| Devstral Small 2 (24B)Mistral AIOpen source | Excellent | Acceptable | 5-bit | 19–25 tok/s |
| Qwen3 32BAlibaba QwenOpen source | Excellent | Acceptable | 4-bit | 17–25 tok/s |
| Qwen2.5-Coder 7B InstructAlibaba QwenOpen source | Excellent | Acceptable | 6-bit | 50–70 tok/s |
| Qwen2.5-Coder 32B InstructAlibaba QwenOpen source | Excellent | Acceptable | 4-bit | 17–25 tok/s |
| DeepSeek-V4-Flash (284B-A13B)DeepSeekOpen source | Comfortable | Acceptable | 4-bit | 35–45 tok/s |
Usable with compromises (1)
Works, but you'll notice slower replies or tight memory.
| Model | Chat | Agentic coding | Quant | Chat speed |
|---|---|---|---|---|
| Llama 3.3 70B InstructMetaOpen weights (restricted) | Acceptable | Borderline | 4-bit | 8–11 tok/s |
Doesn't fit or isn't supported (5)
Not enough usable memory, or the runtime can't run it here.
| Model | Chat | Agentic coding | Quant | Chat speed |
|---|---|---|---|---|
| DeepSeek-V4.1-Flash (748B-A16B)DeepSeekOpen source | Does Not Fit | Does Not Fit | — | — |
| Muse Spark Open 816B-A42BMeta | Does Not Fit | Does Not Fit | — | — |
| Reflection Beam (501B-A23B)Reflection AIOpen source | Does Not Fit | Does Not Fit | — | — |
| Step 5 Preview 600B-A27BStepfun | Does Not Fit | Does Not Fit | — | — |
| GLM-5.3 (753B-A40B)Z.ai | Does Not Fit | Does Not Fit | — | — |
Want a specific tool, context size or workload?
The full check models your exact setup, including the apps you keep open.