Hardware

What can you run?

Curated Q4 VRAM × 1.08 overhead against your usable memory budget — Mac budgets are already unified-memory discounted. See methodology.

Hardware profile

Budget

24 GB

Open in explorer →

ModelParamsQ4 VRAMRequired ×1.08HeadroomVerdictEst. tok/s

SantaCoder 1.1BOPEN1.1B0.8 GB0.9 GBfits comfortably—

Phi-1 (1.3B)OPEN1.3B1 GB1.1 GBfits comfortably—

Phi-1.5 (1.3B)OPEN1.3B1 GB1.1 GBfits comfortably—

Stable LM 2 1.6BOPEN1.64B1.1 GB1.2 GBfits comfortably—

Pythia-1.4BOPEN1.4B1.2 GB1.3 GBfits comfortably—

Qwen-1.8BOPEN1.8B1.2 GB1.3 GBfits comfortably—

Llama 3.2 1BOPEN1.24B1.3 GB1.4 GBfits comfortably—

H2O-Danube 1.8BOPEN1.8B1.4 GB1.5 GBfits comfortably—

SmolLM2 1.7BOPEN1.71B1.5 GB1.6 GBfits comfortably—

Granite 3.0 2B InstructOPEN2.53B1.6 GB1.7 GBfits comfortably—

Phi-2 (2.7B)OPEN2.7B1.7 GB1.8 GBfits comfortably—

Apple OpenELM (270M–3B)OPEN3B1.9 GB2.1 GBfits comfortably—

BTLM-3B-8KOPEN2.6B2 GB2.2 GBfits comfortably—

Gemma 2BOPEN2.5B2 GB2.2 GBfits comfortably—

MiniCPM-2BOPEN2.7B2 GB2.2 GBfits comfortably—

Sarvam-1 (2B)OPEN2B2 GB2.2 GBfits comfortably—

StarCoder2 3BOPEN3B2.1 GB2.3 GBfits comfortably—

GPT-Neo 2.7BOPEN2.7B2.2 GB2.4 GBfits comfortably—

rinna japanese-gpt-neox-3.6bOPEN3.6B2.2 GB2.4 GBfits comfortably—

StableLM-3B-4e1tOPEN2.8B2.2 GB2.4 GBfits comfortably—

StableLM Zephyr 3BOPEN2.8B2.2 GB2.4 GBfits comfortably—

Phi-3-mini (3.8B)OPEN3.8B2.3 GB2.5 GBfits comfortably—

Llama 3.2 3BOPEN3.21B2.5 GB2.7 GBfits comfortably—

SmolLM3 3B (Think)OPEN3.08B2.5 GB2.7 GBfits comfortably—

Gemma 3 4BOPEN4.3B2.6 GB2.8 GBfits comfortably—

H2O-Danube3 4BOPEN3.96B2.6 GB2.8 GBfits comfortably—

MiniCPM3-4BOPEN4B3 GB3.2 GBfits comfortably—

Phi-4-mini (3.8B)OPEN3.8B3 GB3.2 GBfits comfortably—

RedPajama-INCITE-Base-3B-v1OPEN2.8B3 GB3.2 GBfits comfortably—

SmolLM3 3B (No Thinking)OPEN3.08B3 GB3.2 GBfits comfortably~300

Llama-3.1-Minitron 4BOPEN4.5B3.5 GB3.8 GBfits comfortably—

Phi-3.5-mini (3.8B)OPEN3.8B4 GB4.3 GBfits comfortably—

Phi-3.5-vision (4.2B)OPEN4.2B4 GB4.3 GBfits comfortably—

Phi-4-mini-instructOPEN3.8B4 GB4.3 GBfits comfortably—

Phi-4-multimodal (5.6B)OPEN5.6B4 GB4.3 GBfits comfortably—

Llama 2 7B ChatOPEN7B4.1 GB4.4 GBfits comfortably—

WizardLM 7BOPEN7B4.1 GB4.4 GBfits comfortably—

GPT-J-6BOPEN6.05B4.2 GB4.5 GBfits comfortably—

OLMo 7BOPEN6.9B4.2 GB4.5 GBfits comfortably—

OLMoE-1B-7B-0924OPEN6.9B·1.3Ba4.2 GB4.5 GBfits comfortably—

Mistral 7BOPEN7.3B4.4 GB4.8 GBfits comfortably—

MPT-7B-StoryWriter-65k+OPEN6.7B4.4 GB4.8 GBfits comfortably—

Baichuan2-7BOPEN7B4.5 GB4.9 GBfits comfortably—

DeepSeekMath 7BOPEN7B4.5 GB4.9 GBfits comfortably—

InternLM2.5-7BOPEN7B4.5 GB4.9 GBfits comfortably—

MPT-7BOPEN6.7B4.5 GB4.9 GBfits comfortably—

MPT-7B-InstructOPEN6.7B4.5 GB4.9 GBfits comfortably—

OpenLLaMA 7BOPEN6.7B4.5 GB4.9 GBfits comfortably—

Pharia-1-LLM-7B (control)OPEN7.04B4.5 GB4.9 GBfits comfortably—

Qwen-7BOPEN7B4.5 GB4.9 GBfits comfortably—

StableLM-Alpha 7BOPEN7B4.5 GB4.9 GBfits comfortably—

Starling-7B (alpha)OPEN7B4.5 GB4.9 GBfits comfortably~135

Yi-6BOPEN6B4.5 GB4.9 GBfits comfortably—

Falcon3-7BOPEN7B4.6 GB5.0 GBfits comfortably—

Granite 8B Code InstructOPEN8B4.6 GB5.0 GBfits comfortably—

Falcon-H1R 7BOPEN7.6B4.7 GB5.1 GBfits comfortably—

Pythia-6.9BOPEN6.9B4.7 GB5.1 GBfits comfortably—

Qwen2.5-7BOPEN7.61B4.7 GB5.1 GBfits comfortably—

Skywork-o1-Open-Llama-3.1-8BOPEN8B4.92 GB5.3 GBfits comfortably—

BLOOM-7B1OPEN7.069B5 GB5.4 GBfits comfortably—

Code Llama 7BOPEN7B5 GB5.4 GBfits comfortably—

CodeGen2.5 7BOPEN7B5 GB5.4 GBfits comfortably—

CodeQwen1.5-7BOPEN7B5 GB5.4 GBfits comfortably—

Codestral Mamba (7B)OPEN7.3B5 GB5.4 GBfits comfortably—

Command R7BOPEN7B5 GB5.4 GBfits comfortably—

DeciLM-7BOPEN7.04B5 GB5.4 GBfits comfortably—

DeepSeek-VL 7BOPEN7B5 GB5.4 GBfits comfortably—

Falcon Mamba 7BOPEN7B5 GB5.4 GBfits comfortably—

Gorilla 7BOPEN7B5 GB5.4 GBfits comfortably—

LG EXAONE 3.0 7.8B InstructOPEN7.8B5 GB5.4 GBfits comfortably—

Llama-3-ELYZA-JP-8BOPEN8B5 GB5.4 GBfits comfortably—

Ministral 8BOPEN8.02B5 GB5.4 GBfits comfortably—

Molmo 2 8BOPEN8.66B5 GB5.4 GBfits comfortably—

OLMo 2 7BOPEN7.3B5 GB5.4 GBfits comfortably—

OLMo 3 7BOPEN7.3B5 GB5.4 GBfits comfortably—

Phi-4-multimodal-instructOPEN5.6B5 GB5.4 GBfits comfortably—

Qwen2-7BOPEN7.07B5 GB5.4 GBfits comfortably—

RedPajama-INCITE-7B-BaseOPEN6.9B5 GB5.4 GBfits comfortably—

RedPajama-INCITE-7B-ChatOPEN6.9B5 GB5.4 GBfits comfortably—

RedPajama-INCITE-7B-InstructOPEN6.9B5 GB5.4 GBfits comfortably—

Teuken-7B (instruct)OPEN7B5 GB5.4 GBfits comfortably—

Viking 7BOPEN7.55B5 GB5.4 GBfits comfortably—

Zephyr 7B BetaOPEN7.24B5 GB5.4 GBfits comfortably—

Qwen3-8B (Non-Thinking)OPEN8.2B5.1 GB5.5 GBfits comfortably—

Qwen3-8B (Thinking)OPEN8.2B5.1 GB5.5 GBfits comfortably—

Nemotron-3 8BOPEN8B5.4 GB5.8 GBfits comfortably—

ChatGLM2-6BOPEN6.24B5.5 GB5.9 GBfits comfortably—

InternLM3-8B-InstructOPEN8B5.5 GB5.9 GBfits comfortably—

Llama 3.1 8BOPEN8.03B5.5 GB5.9 GBfits comfortably~100

Llama 3 8BOPEN8.03B5.5 GB5.9 GBfits comfortably~100

Alpaca 7BOPEN6.7B6 GB6.5 GBfits comfortably—

ChatGLM-6BOPEN6.2B6 GB6.5 GBfits comfortably—

ChatGLM3-6BOPEN6.2B6 GB6.5 GBfits comfortably—

CodeGemma 7BOPEN8.54B6 GB6.5 GBfits comfortably—

Falcon-7BOPEN7B6 GB6.5 GBfits comfortably—

Falcon3-10BOPEN10B6 GB6.5 GBfits comfortably—

Gemma 7BOPEN8.54B6 GB6.5 GBfits comfortably—

GLM-4-9B-ChatOPEN9B6 GB6.5 GBfits comfortably—

Granite 3.0 8B InstructOPEN8.1B6 GB6.5 GBfits comfortably—

Granite 3.1 8B InstructOPEN8.1B6 GB6.5 GBfits comfortably—

Granite 3.2 8B InstructOPEN8.1B6 GB6.5 GBfits comfortably—

Granite 3.3 8B InstructOPEN8.1B6 GB6.5 GBfits comfortably—

Idefics2 8BOPEN8B6 GB6.5 GBfits comfortably—

MiniCPM-Llama3-V 2.5OPEN8B6 GB6.5 GBfits comfortably—

MiniCPM-V 2.6OPEN8B6 GB6.5 GBfits comfortably—

OpenHermes 2.5 Mistral 7BOPEN7.24B6 GB6.5 GBfits comfortably—

Phi-3-small (7B)OPEN7B6 GB6.5 GBfits comfortably—

Qwen-VLOPEN9.6B6 GB6.5 GBfits comfortably—

Yi-1.5 9BOPEN9B6 GB6.5 GBfits comfortably—

Gemma 3 12BOPEN12.2B6.6 GB7.1 GBfits comfortably—

Llama 2 7BOPEN6.7B6.6 GB7.1 GBfits comfortably—

LLaMA 7BOPEN6.7B6.6 GB7.1 GBfits comfortably—

Baichuan2-13BOPEN13B7 GB7.6 GBfits comfortably—

Cerebras-GPT 13BOPEN13B7 GB7.6 GBfits comfortably—

Falcon 2 11BOPEN11B7 GB7.6 GBfits comfortably—

Flan-T5 XXLOPEN11B7 GB7.6 GBfits comfortably—

Gemma 2 9BOPEN9.24B7 GB7.6 GBfits comfortably—

MiniCPM-o 2.6OPEN8B7 GB7.6 GBfits comfortably—

Mistral NeMo 12BOPEN12B7 GB7.6 GBfits comfortably—

Upstage SOLAR 10.7BOPEN10.7B7 GB7.6 GBfits comfortably—

Krutrim-2 (12B)OPEN12B7.5 GB8.1 GBfits comfortably—

Llama 3.2 11B VisionOPEN10.6B7.5 GB8.1 GBfits comfortably—

Orca 2 (13B)OPEN13B7.5 GB8.1 GBfits comfortably—

Pixtral 12BOPEN12.4B7.5 GB8.1 GBfits comfortably—

Pythia-12BOPEN11.8B7.9 GB8.5 GBfits comfortably—

Code Llama 13BOPEN13B8 GB8.6 GBfits comfortably—

GLM-4V-9BOPEN14B8 GB8.6 GBfits comfortably—

Koala 13BOPEN13B8 GB8.6 GBfits comfortably—

LLaMA 13BOPEN13B8 GB8.6 GBfits comfortably—

Nemotron Nano 2 (Nemotron-Nano-9B-v2)OPEN9B8 GB8.6 GBfits comfortably—

Stable LM 2 12BOPEN12.1B8 GB8.6 GBfits comfortably—

OLMo 2 13BOPEN13B8.4 GB9.1 GBfits comfortably—

Qwen3-14B (Non-Thinking)OPEN14.8B8.4 GB9.1 GBfits comfortably~60

Qwen3-14B (Thinking)OPEN14.8B8.4 GB9.1 GBfits comfortably~60

Skywork-13BOPEN13B8.4 GB9.1 GBfits comfortably—

Jais-13BOPEN13B9 GB9.7 GBfits comfortably—

Phi-3-medium (14B)OPEN14B9 GB9.7 GBfits comfortably—

Qwen-14BOPEN14B9 GB9.7 GBfits comfortably—

Qwen1.5-MoE-A2.7BOPEN14.3B·2.7Ba9 GB9.7 GBfits comfortably—

Qwen2.5-14BOPEN14.7B9 GB9.7 GBfits comfortably—

Vicuna 13B (v1.1)OPEN13B9 GB9.7 GBfits comfortably—

Vicuna 13B v1.5OPEN13B9 GB9.7 GBfits comfortably—

StarCoder2 15BOPEN15B9.4 GB10.2 GBfits comfortably—

StarCoder (15.5B)OPEN15.5B9.7 GB10.5 GBfits comfortably—

DeepSeekMoE 16BOPEN16.4B·2.8Ba10 GB10.8 GBfits comfortably—

Llama 2 13BOPEN13B10 GB10.8 GBfits comfortably—

Llama 2 13B ChatOPEN13B10 GB10.8 GBfits comfortably~70.9

Nemotron-4 15BOPEN15B10 GB10.8 GBfits comfortably—

Phi-4-reasoning (14B)OPEN14B10 GB10.8 GBfits comfortably—

CogVLM-17BOPEN17B11 GB11.9 GBfits comfortably—

Phi-4 (14B)OPEN14B11 GB11.9 GBfits comfortably—

InternLM2-20BOPEN20B12 GB13.0 GBfits comfortably—

gpt-oss-20b (High)OPEN21B·3.6Ba13 GB14.0 GBfits comfortably~225

gpt-oss-20b (Low)OPEN21B·3.6Ba13 GB14.0 GBfits comfortably~225

gpt-oss-20b (Medium)OPEN21B·3.6Ba13 GB14.0 GBfits comfortably~225

Upstage Solar ProOPEN22B13 GB14.0 GBfits comfortably—

GPT-NeoX-20BOPEN20.6B13.1 GB14.1 GBfits comfortably—

Phi-4-reasoningOPEN14B13.8 GB14.9 GBfits comfortably—

DeepSeek-VL2OPEN27B·4.5Ba14 GB15.1 GBfits comfortably—

Reka Flash 3OPEN21B14 GB15.1 GBfits comfortably—

Gemma 3 27BOPEN27.4B14.1 GB15.2 GBfits comfortably—

Devstral SmallOPEN24B14.3 GB15.4 GBfits comfortably—

Magistral SmallOPEN24B14.5 GB15.7 GBfits comfortably—

Mistral Small 3.2 (24B)OPEN24B14.5 GB15.7 GBfits comfortably~70

Mistral Small 3 (24B)OPEN24B14.5 GB15.7 GBfits comfortably—

Mistral Small 3.1 (24B)OPEN24B15 GB16.2 GBfits comfortably—

Gemma 2 27BOPEN27.2B16 GB17.3 GBfits comfortably—

OPT-IML 30BOPEN30B17 GB18.4 GBfits comfortably—

Codestral 22BOPEN22.2B18 GB19.4 GBfits (tight)—

Qwen3-30B-A3B (Non-Thinking)OPEN30.5B·3.3Ba18.6 GB20.1 GBfits (tight)—

Qwen3-30B-A3B (Thinking)OPEN30.5B·3.3Ba18.6 GB20.1 GBfits (tight)—

Falcon-H1 34BOPEN34B19 GB20.5 GBfits (tight)—

LLaMA 33BOPEN32.5B19 GB20.5 GBfits (tight)—

MPT-30BOPEN30B19.1 GB20.6 GBfits (tight)—

MPT-30B-InstructOPEN30B19.1 GB20.6 GBfits (tight)—

QwQ-32BOPEN32.5B19.85 GB21.4 GBfits (tight)—

QwQ-32B-PreviewOPEN32.5B19.85 GB21.4 GBfits (tight)—

Qwen2.5-32BOPEN32.5B19.9 GB21.5 GBfits (tight)—

Qwen2.5-Coder-32BOPEN32.5B19.9 GB21.5 GBfits (tight)—

Qwen3-32B (Non-Thinking)OPEN32.8B19.9 GB21.5 GBfits (tight)—

Qwen3-32B (Thinking)OPEN32.8B19.9 GB21.5 GBfits (tight)—

Aya Expanse 32BOPEN32B20 GB21.6 GBfits (tight)—

DeepSeek Coder 33BOPEN33.3B20 GB21.6 GBfits (tight)—

DeepSeek-R1-Distill-Qwen-32BOPEN32B20 GB21.6 GBfits (tight)~40

Gemma 4OPEN30.7B20 GB21.6 GBfits (tight)—

GLM-4-32B-0414OPEN32B20 GB21.6 GBfits (tight)—

GLM-Z1-32B-0414OPEN32B20 GB21.6 GBfits (tight)—

LG EXAONE 3.5 32B InstructOPEN32B20 GB21.6 GBfits (tight)—

LG EXAONE 4.0 32B (Non-reasoning)OPEN32B20 GB21.6 GBfits (tight)—

LG EXAONE 4.0 32B (Reasoning)OPEN32B20 GB21.6 GBfits (tight)—

LG EXAONE Deep 32BOPEN32B20 GB21.6 GBfits (tight)—

OLMo 2 32BOPEN32.23B20 GB21.6 GBfits (tight)—

OLMo 3-Think 32BOPEN32.23B20 GB21.6 GBfits (tight)—

Poro 34BOPEN34.2B20 GB21.6 GBfits (tight)—

WizardCoder-Python-34B V1.0OPEN34B20 GB21.6 GBfits (tight)—

Yi-1.5 34BOPEN34.4B20 GB21.6 GBfits (tight)—

Granite 4.0 H SmallOPEN32B·9Ba21 GB22.7 GBfits (tight)—

Guanaco 33BOPEN33B21 GB22.7 GBfits (tight)—

Granite 34B Code InstructOPEN34B21.4 GB23.1 GBfits (tight)—

Command ROPEN35B21.6 GB23.3 GBfits (tight)—

Seed-OSS-36BOPEN36B21.8 GB23.5 GBfits (tight)—

Aya 23 35BOPEN35B22 GB23.8 GBfits (tight)—

Code Llama 34BOPEN34B22 GB23.8 GBfits (tight)—

Code Llama 34B PythonOPEN34B22 GB23.8 GBfits (tight)—

Yi-34BOPEN34B22 GB23.8 GBfits (tight)—

Falcon-40BOPEN40B23 GB24.8 GBpartial offload—

Dolphin 2.5 Mixtral 8x7BOPEN46.7B·12.9Ba26 GB28.1 GBpartial offload—

Mixtral 8x7BOPEN46.7B·12.9Ba26 GB28.1 GBpartial offload—

Phi-3.5-MoE (16x3.8B, 6.6B active)OPEN41.9B·6.6Ba27 GB29.2 GBpartial offload—

AI21 JambaOPEN52B·12Ba29.5 GB31.9 GBwon't run—

AI21 Jamba 1.5 MiniOPEN52B·12Ba29.5 GB31.9 GBwon't run—

Llama-3.3-Nemotron-Super-49B v1 (Reasoning Off)OPEN49.9B30 GB32.4 GBwon't run—

Llama-3.3-Nemotron-Super-49B v1 (Reasoning On)OPEN49.9B30 GB32.4 GBwon't run—

Nous Hermes 2 Mixtral 8x7B DPOOPEN46.7B·12.9Ba30 GB32.4 GBwon't run—

Qwen2-57B-A14BOPEN57B·14Ba32 GB34.6 GBwon't run—

OPT-66BOPEN66B35 GB37.8 GBwon't run—

LLaMA 65BOPEN65.2B37 GB40.0 GBwon't run—

Llama 2 70BOPEN69B38 GB41.0 GBwon't run—

Llama 2 70B ChatOPEN69B38 GB41.0 GBwon't run—

DeepSeek LLM 67BOPEN67B39 GB42.1 GBwon't run—

Guanaco 65BOPEN65B39 GB42.1 GBwon't run—

Molmo 72BOPEN72B39 GB42.1 GBwon't run—

Airoboros 70B (L2)OPEN70B40 GB43.2 GBwon't run—

Code Llama 70BOPEN70B40 GB43.2 GBwon't run~18

Code Llama 70B InstructOPEN70B40 GB43.2 GBwon't run~18

Jais 70BOPEN70B40 GB43.2 GBwon't run—

Llama 3.1 70BOPEN70.6B40 GB43.2 GBwon't run—

Llama 3 70BOPEN70B40 GB43.2 GBwon't run~18

Sarashina2-70BOPEN70B40 GB43.2 GBwon't run—

Stable Beluga 2 (FreeWilly 2) 70BOPEN70B40 GB43.2 GBwon't run—

Tülu 2+DPO 70BOPEN69B41.4 GB44.7 GBwon't run—

DeepSeek-R1-Distill-Llama-70BOPEN70B42 GB45.4 GBwon't run—

Llama-3.1-Nemotron-70B InstructOPEN70.6B42 GB45.4 GBwon't run—

Llama 3.3 70BOPEN70B42 GB45.4 GBwon't run—

Palmyra Med 70BOPEN70B42 GB45.4 GBwon't run—

Qwen2.5-VL-72BOPEN73B42 GB45.4 GBwon't run—

Qwen2-72BOPEN72B42 GB45.4 GBwon't run—

Qwen2-Math-72BOPEN72.7B42 GB45.4 GBwon't run—

Tülu 3 70BOPEN70B42.5 GB45.9 GBwon't run—

Hunyuan-A13BOPEN80B·13Ba44 GB47.5 GBwon't run—

QVQ-72B-PreviewOPEN73B47 GB50.8 GBwon't run—

Qwen2.5-Math-72BOPEN72.7B47 GB50.8 GBwon't run—

Qwen2.5-72BOPEN72.7B47.4 GB51.2 GBwon't run—

Qwen-72BOPEN72B48 GB51.8 GBwon't run—

Qwen1.5-72BOPEN72B48 GB51.8 GBwon't run—

Qwen2-VL-72BOPEN73.4B48 GB51.8 GBwon't run—

InternVL2.5-78BOPEN78B53 GB57.2 GBwon't run—

InternVL3-78BOPEN78B53 GB57.2 GBwon't run—

GLM-4.5-Air (106B-A12B)OPEN106B·12Ba55 GB59.4 GBwon't run—

Llama 3.2 90B VisionOPEN88.8B55 GB59.4 GBwon't run—

Llama 4 ScoutOPEN109B·17Ba58 GB62.6 GBwon't run—

Galactica 120BOPEN120B64 GB69.1 GBwon't run—

Qwen1.5-110BOPEN111B65 GB70.2 GBwon't run—

Sarvam-105BOPEN105B·10.3Ba65 GB70.2 GBwon't run—

DBRXOPEN132B·36Ba66 GB71.3 GBwon't run—

Command AOPEN111B67 GB72.4 GBwon't run—

gpt-oss-120b (High)OPEN117B·5.1Ba73 GB78.8 GBwon't run—

gpt-oss-120b (Low)OPEN117B·5.1Ba73 GB78.8 GBwon't run—

gpt-oss-120b (Medium)OPEN117B·5.1Ba73 GB78.8 GBwon't run—

Mistral Large 2 (24.07)OPEN123B74 GB79.9 GBwon't run—

Mistral Large 24.11OPEN123B74 GB79.9 GBwon't run—

Pixtral Large (124B)OPEN124B75 GB81.0 GBwon't run—

Mixtral 8x22BOPEN141B·39Ba80 GB86.4 GBwon't run—

WizardLM-2 8x22BOPEN141B·39Ba80 GB86.4 GBwon't run—

OPT-175BOPEN175B90 GB97.2 GBwon't run—

GLM-130BOPEN130B96 GB103.7 GBwon't run—

Falcon-180BOPEN180B100 GB108.0 GBwon't run—

BLOOM (176B)OPEN176.25B106 GB114.5 GBwon't run—

BLOOMZ (176B)OPEN176.25B106 GB114.5 GBwon't run—

DeepSeek-Coder-V2 236BOPEN236B·21Ba118 GB127.4 GBwon't run—

DeepSeek-V2 236BOPEN236B·21Ba118 GB127.4 GBwon't run—

DeepSeek-V2.5OPEN236B·21Ba118 GB127.4 GBwon't run—

Llama-3.1-Nemotron-Ultra-253B v1 (Reasoning Off)OPEN253B126.5 GB136.6 GBwon't run—

Llama-3.1-Nemotron-Ultra-253B v1 (Reasoning On)OPEN253B126.5 GB136.6 GBwon't run—

MiniMax-M2OPEN229.9B·9.8Ba130 GB140.4 GBwon't run—

Qwen3-235B-A22B (Non-Thinking)OPEN235B·22Ba132 GB142.6 GBwon't run—

Qwen3-235B-A22B (Thinking)OPEN235B·22Ba132 GB142.6 GBwon't run—

Qwen3-235B-A22B-Instruct-2507OPEN235B·22Ba142 GB153.4 GBwon't run—

ERNIE 4.5 300B-A47BOPEN300B·47Ba151 GB163.1 GBwon't run—

Step-3OPEN321B·38Ba177 GB191.2 GBwon't run—

GLM-4.5 (355B-A32B)OPEN355B·32Ba180 GB194.4 GBwon't run—

Grok-1OPEN314B·86Ba180 GB194.4 GBwon't run—

Nemotron-4 340BOPEN340B200 GB216.0 GBwon't run—

Llama 4 MaverickOPEN400B·17Ba210 GB226.8 GBwon't run—

Hunyuan-Large (A52B)OPEN389B·52Ba215 GB232.2 GBwon't run—

Llama 3.1 405BOPEN405B230 GB248.4 GBwon't run—

Tülu 3 405BOPEN405.85B230 GB248.4 GBwon't run—

AI21 Jamba 1.5 LargeOPEN398B·94Ba240.4 GB259.6 GBwon't run—

ERNIE 4.5 VL 424B-A47BOPEN424B·47Ba245 GB264.6 GBwon't run—

Hermes 3 405BOPEN405B250 GB270.0 GBwon't run—

Qwen3-Coder-480B-A35BOPEN480B·35Ba289 GB312.1 GBwon't run—

Snowflake ArcticOPEN480B·17Ba291 GB314.3 GBwon't run—

R1-1776OPEN671B·37Ba335 GB361.8 GBwon't run—

Nemotron 3 Ultra 550B A55BOPEN550B·55Ba352 GB380.2 GBwon't run—

DeepSeek-V3 671BOPEN671B·37Ba400 GB432.0 GBwon't run—

DeepSeek-Prover-V2 671BOPEN671B·37Ba404 GB436.3 GBwon't run—

DeepSeek-R1-0528OPEN671B·37Ba404 GB436.3 GBwon't run—

DeepSeek-R1 671BOPEN671B·37Ba404 GB436.3 GBwon't run—

DeepSeek-R1-ZeroOPEN671B·37Ba404 GB436.3 GBwon't run—

DeepSeek-V3-0324OPEN671B·37Ba404 GB436.3 GBwon't run—

DeepSeek-V3.1 (Non-Thinking)OPEN671B·37Ba405 GB437.4 GBwon't run—

DeepSeek-V3.1 (Thinking)OPEN671B·37Ba405 GB437.4 GBwon't run—

InklingOPEN975B·41Ba600 GB648.0 GBwon't run—

Kimi K2OPEN1000B·32Ba600 GB648.0 GBwon't run—

Kimi K2 ThinkingOPEN1000B·32Ba600 GB648.0 GBwon't run—

tok/s shown only where a measured throughput row exists (RTX 4090 · Q4 · llama.cpp). MoE models need total-parameter memory; speed tracks active parameters.