Comfortable
Qwen 2.5 0.5B (Q4)
Large language models (LLM)
Lightest model, for embedded devices
Minimum 1GB / Recommended 2GB
Comfortable
Qwen 2.5 1.5B (Q4)
Large language models (LLM)
Ultra-lightweight, runs on phones
Minimum 2GB / Recommended 3GB
Comfortable
Phi-3 mini 3.8B (Q4)
Large language models (LLM)
Small, high-performance model by Microsoft
Minimum 3GB / Recommended 4GB
Comfortable
Gemma 3 4B (Q4)
Large language models (LLM)
Lightweight model by Google
Minimum 3GB / Recommended 4GB
Comfortable
Llama 3.2 3B (Q4)
Large language models (LLM)
Small mobile model by Meta
Minimum 3GB / Recommended 4GB
Comfortable
Llama 3.2 1B (Q4)
Large language models (LLM)
Smallest mobile model by Meta
Minimum 2GB / Recommended 3GB
Comfortable
Mistral 7B (Q2)
Large language models (LLM)
Extreme quantization, quality trade-off
Minimum 3GB / Recommended 4GB
Comfortable
Mistral 7B (Q4)
Large language models (LLM)
Efficient, recommended setting
Minimum 5GB / Recommended 6GB
Comfortable
Mistral 7B (Q5)
Large language models (LLM)
Balanced quantization
Minimum 6GB / Recommended 7GB
Comfortable
Mistral 7B (Q8)
Large language models (LLM)
High-quality quantization
Minimum 8GB / Recommended 10GB
Comfortable
Llama 3.1 8B (Q4)
Large language models (LLM)
Meta, standard for chat
Minimum 6GB / Recommended 8GB
Comfortable
Llama 3.1 8B (Q5)
Large language models (LLM)
Balanced quantization
Minimum 7GB / Recommended 9GB
Comfortable
Llama 3.1 8B (Q8)
Large language models (LLM)
Near full quality
Minimum 10GB / Recommended 12GB
Comfortable
Llama 3.1 8B (BF16)
Large language models (LLM)
Full precision
Minimum 16GB / Recommended 18GB
Comfortable
Qwen 2.5 7B (Q4)
Large language models (LLM)
Strong in Japanese, by Alibaba
Minimum 5GB / Recommended 7GB
Comfortable
Gemma 2 9B (Q4)
Large language models (LLM)
Mid-size model by Google
Minimum 6GB / Recommended 8GB
Comfortable
Phi-4 14B (Q4)
Large language models (LLM)
Compact, high-performance model by Microsoft
Minimum 9GB / Recommended 12GB
Comfortable
Qwen 2.5 14B (Q4)
Large language models (LLM)
Mid-size model by Alibaba
Minimum 9GB / Recommended 12GB
Comfortable
Qwen 2.5 32B (Q4)
Large language models (LLM)
Mid-scale, high performance
Minimum 19GB / Recommended 24GB
Comfortable
Llama 3.3 70B (Q2)
Large language models (LLM)
Extreme quantization, quality trade-off
Minimum 28GB / Recommended 32GB
Comfortable
Llama 3.3 70B (Q4)
Large language models (LLM)
GPT-4 class, large scale
Minimum 40GB / Recommended 48GB
Comfortable
Llama 3.3 70B (Q8)
Large language models (LLM)
Near full precision
Minimum 70GB / Recommended 80GB
Comfortable
Mixtral 8x7B (Q4)
Large language models (LLM)
MoE architecture, high performance
Minimum 24GB / Recommended 32GB
Comfortable
Mixtral 8x22B (Q4)
Large language models (LLM)
Very large MoE
Minimum 80GB / Recommended 96GB
Comfortable
DeepSeek-R1 Distill 7B (Q4)
Large language models (LLM)
Specialized for reasoning, lightweight
Minimum 5GB / Recommended 7GB
Comfortable
DeepSeek-R1 Distill 32B (Q4)
Large language models (LLM)
Specialized for reasoning, mid-size
Minimum 19GB / Recommended 24GB
Comfortable
Command-R 35B (Q4)
Large language models (LLM)
By Cohere, specialized for RAG
Minimum 20GB / Recommended 24GB
Comfortable
Command-R+ 104B (Q4)
Large language models (LLM)
Cohere's most capable model, commercial use
Minimum 60GB / Recommended 80GB
Comfortable
Qwen 2.5 72B (Q4)
Large language models (LLM)
Among the strongest open models
Minimum 40GB / Recommended 48GB
Comfortable
ELYZA-Llama 8B (Q4)
Large language models (LLM)
Llama specialized for Japanese
Minimum 6GB / Recommended 8GB
Comfortable
Swallow 70B (Q4)
Large language models (LLM)
Enhanced for Japanese, Institute of Science Tokyo
Minimum 40GB / Recommended 48GB
Comfortable
LLaVA 1.6 7B (Q4)
Large language models (LLM)
Multimodal model for image understanding
Minimum 7GB / Recommended 9GB
Comfortable
Qwen2-VL 7B (Q4)
Large language models (LLM)
Video and image understanding
Minimum 7GB / Recommended 9GB
Comfortable
Qwen3 8B (Q4)
Large language models (LLM)
2025 generation, efficient 8B with switchable thinking mode
Minimum 5GB / Recommended 6GB
Comfortable
Qwen3 14B (Q4)
Large language models (LLM)
2025 generation, balanced 14B
Minimum 8GB / Recommended 10GB
Comfortable
Qwen3 30B-A3B (Q4)
Large language models (LLM)
MoE. 30B-class quality at 3B-class speed
Minimum 20GB / Recommended 24GB
Comfortable
Qwen3 32B (Q4)
Large language models (LLM)
Delivers Qwen2.5 72B-class performance at 32B
Minimum 19GB / Recommended 24GB
Comfortable
Qwen3.5 9B (Q4)
Large language models (LLM)
February 2026 version, 262K long context
Minimum 6GB / Recommended 8GB
Comfortable
Qwen3.5 27B (Q4)
Large language models (LLM)
February 2026 version, dense 27B
Minimum 18GB / Recommended 24GB
Comfortable
Qwen3.5 35B-A3B (Q4)
Large language models (LLM)
February 2026, MoE with 35B total and 3B active, 262K long context
Minimum 22GB / Recommended 24GB
Comfortable
Qwen3.6 27B (Q4)
Large language models (LLM)
April 2026, dense 27B. Among the strongest for coding, 1M context
Minimum 18GB / Recommended 24GB
Comfortable
Qwen3.6 35B-A3B (Q4)
Large language models (LLM)
April 2026, latest MoE with 35B total and 3B active. Specialized for agents and coding
Minimum 22GB / Recommended 24GB
Comfortable
Gemma 3 12B (Q4)
Large language models (LLM)
Google Gemma 3, mid-size, accepts image input
Minimum 7GB / Recommended 9GB
Comfortable
Gemma 3 27B (Q4)
Large language models (LLM)
Google Gemma 3, largest 27B model, multimodal
Minimum 18GB / Recommended 24GB
Comfortable
Gemma 4 E2B (Q4)
Large language models (LLM)
Google, April 2026, smallest phone-class model, accepts image and audio input
Minimum 4GB / Recommended 6GB
Comfortable
Qwen3.5 2B (Q4)
Large language models (LLM)
Alibaba, 2026, small multimodal model with thinking mode
Minimum 4GB / Recommended 6GB
Comfortable
Gemma 4 E4B (Q4)
Large language models (LLM)
Google, April 2026, lightweight multimodal, Apache 2.0
Minimum 6GB / Recommended 8GB
Comfortable
Qwen3.5 4B (Q4)
Large language models (LLM)
Alibaba, 2026, high performance for 4B class, 262K context
Minimum 6GB / Recommended 8GB
Comfortable
Gemma 4 12B (Q4)
Large language models (LLM)
Google, added June 2026, mid-size model for image and audio
Minimum 8GB / Recommended 10GB
Comfortable
Ministral 3 14B (Q4)
Large language models (LLM)
Mistral, December 2025, 14B, a reasoning version is also available
Minimum 10GB / Recommended 12GB
Comfortable
Ministral 3 8B (FP8)
Large language models (LLM)
Mistral, December 2025, supports Japanese and image input, official FP8 needs 12GB
Minimum 12GB / Recommended 16GB
Comfortable
gpt-oss-20b (MXFP4)
Large language models (LLM)
OpenAI's first open model, MoE with 21B total and 3.6B active, official quantization that runs in 16GB, Apache 2.0
Minimum 16GB / Recommended 20GB
Comfortable
GPT-OSS Swallow 20B (MXFP4)
Large language models (LLM)
Institute of Science Tokyo and AIST, Japanese-enhanced gpt-oss-20b, Apache 2.0
Minimum 16GB / Recommended 20GB
Comfortable
Mistral Small 3.2 24B (Q4)
Large language models (LLM)
Mistral, general-purpose 24B with image input, Apache 2.0
Minimum 16GB / Recommended 24GB
Comfortable
Devstral Small 2 24B (Q4)
Large language models (LLM)
Mistral, specialized for coding agents, runs on one RTX 4090
Minimum 16GB / Recommended 24GB
Comfortable
Gemma 4 26B A4B (Q4)
Large language models (LLM)
Google, MoE with 26B total and 4B active, focused on speed
Minimum 18GB / Recommended 24GB
Comfortable
Qwen3.8 27B (Q4)
Large language models (LLM)
Alibaba, August 2026, dense 27B that reads images and video, Apache 2.0
Minimum 18GB / Recommended 24GB
Comfortable
Qwen3 Swallow 8B (BF16)
Large language models (LLM)
Institute of Science Tokyo and AIST, Japanese-enhanced Qwen3, vLLM recommended
Minimum 18GB / Recommended 24GB
Comfortable
Gemma 4 31B (Q4)
Large language models (LLM)
Google Gemma 4's largest dense model. Memory grows with long text due to the KV cache
Minimum 20GB / Recommended 24GB
Comfortable
GLM-4.7-Flash (Q4)
Large language models (LLM)
Z.ai, MoE 30B with about 3B active, local coding
Minimum 20GB / Recommended 24GB
Comfortable
Muse Glimmer 30B (Q4)
Large language models (LLM)
Meta, August 2026, 30B for agents, accepts image input, Apache 2.0
Minimum 24GB / Recommended 32GB
Comfortable
Nemotron 3 Nano 30B-A3B (Q4)
Large language models (LLM)
NVIDIA, MoE with Mamba hybrid layers, 1M context
Minimum 24GB / Recommended 32GB
Comfortable
Qwen3.5 122B-A10B (Q4)
Large language models (LLM)
Alibaba, MoE with 122B total and 10B active, for workstations
Minimum 72GB / Recommended 96GB
Comfortable
Llama 4 Scout (Q4)
Large language models (LLM)
Meta, MoE with 109B total and 17B active, accepts image input
Minimum 72GB / Recommended 96GB
Comfortable
gpt-oss-120b (MXFP4)
Large language models (LLM)
OpenAI, MoE with 117B total and 5.1B active, official quantization that runs on one 80GB GPU
Minimum 80GB / Recommended 96GB
Comfortable
Mistral Small 4 119B (Q4)
Large language models (LLM)
Mistral, March 2026, MoE with 119B total and 6B active, combines reasoning, images, and code
Minimum 80GB / Recommended 96GB
Comfortable
llm-jp-4.1 8B thinking (Q4)
Large language models (LLM)
LLM-jp, Sep 2026, Japanese reasoning model, 8B, Apache 2.0
Minimum 7GB / Recommended 10GB
Comfortable
llm-jp-4.1 33B thinking (Q4)
Large language models (LLM)
LLM-jp, Sep 2026, Japanese reasoning model, 33B, Apache 2.0
Minimum 22GB / Recommended 24GB
Comfortable
llm-jp-4.1 32B-A3B thinking (Q4)
Large language models (LLM)
LLM-jp, Sep 2026, Japanese reasoning MoE with 32B total and 3B active, Apache 2.0
Minimum 24GB / Recommended 32GB
Comfortable
ELYZA-Thinking-1.0 33B (Q4)
Large language models (LLM)
ELYZA, Oct 2026, Japanese reasoning model built on llm-jp-4, Apache 2.0
Minimum 22GB / Recommended 24GB
Comfortable
ELYZA-Thinking-1.0 32B-A3B (Q4)
Large language models (LLM)
ELYZA, Oct 2026, Japanese reasoning MoE with 32B total and 3B active, Apache 2.0
Minimum 23GB / Recommended 32GB
Comfortable
llm-jp-4 VL 9B (Q4)
Large language models (LLM)
LLM-jp, Aug 2026, Japanese 9B that can read images, Apache 2.0
Minimum 7GB / Recommended 10GB
Comfortable
MiniCPM5 2B (Q4)
Large language models (LLM)
OpenBMB, Sep 2026, small 2.5B model with long context and tool calling, Apache 2.0
Minimum 4GB / Recommended 6GB
Comfortable
Qwen3.8-Flash-Next (Q4)
Large language models (LLM)
Alibaba, Aug 2026, MoE with 180B total and 6B active, image input, for workstations
Minimum 113GB / Recommended 128GB
Comfortable
Ling-3.0-flash-VL (Q4)
Large language models (LLM)
inclusionAI, Sep 2026, MoE with 124B total and 5.5B active, reads images and video, MIT
Minimum 81GB / Recommended 96GB
Comfortable
Qwen3.8 27B (Q3)
Large language models (LLM)
UD-Q3_K_XL, for running on 16GB cards
Minimum 15GB / Recommended 20GB
Comfortable
Qwen3.8 27B (Q6)
Large language models (LLM)
UD-Q6_K, higher quality than Q4
Minimum 24GB / Recommended 32GB
Comfortable
Qwen3.8 27B (Q8)
Large language models (LLM)
Near full quality
Minimum 31GB / Recommended 48GB
Comfortable
Qwen3.6 35B-A3B (Q3)
Large language models (LLM)
UD-Q3_K_M, for 20GB cards
Minimum 19GB / Recommended 20GB
Comfortable
Qwen3.6 35B-A3B (Q8)
Large language models (LLM)
Near full quality
Minimum 39GB / Recommended 48GB
Comfortable
Qwen3.6 27B (Q3)
Large language models (LLM)
For running on 16GB cards
Minimum 16GB / Recommended 20GB
Comfortable
Qwen3.6 27B (Q8)
Large language models (LLM)
Near full quality
Minimum 31GB / Recommended 32GB
Comfortable
Qwen3.5 9B (Q8)
Large language models (LLM)
Near full quality
Minimum 12GB / Recommended 16GB
Comfortable
Qwen3.5 9B (BF16)
Large language models (LLM)
Full precision
Minimum 20GB / Recommended 24GB
Comfortable
Gemma 4 31B (Q3)
Large language models (LLM)
For running on 16-20GB cards
Minimum 17GB / Recommended 20GB
Comfortable
Gemma 4 31B (Q8)
Large language models (LLM)
Near full quality
Minimum 35GB / Recommended 48GB
Comfortable
Gemma 4 26B A4B (Q3)
Large language models (LLM)
UD-Q3_K_M, for 16GB cards
Minimum 15GB / Recommended 16GB
Comfortable
Gemma 4 26B A4B (Q8)
Large language models (LLM)
Near full quality
Minimum 29GB / Recommended 32GB
Comfortable
Gemma 4 12B (Q8)
Large language models (LLM)
Near full quality
Minimum 15GB / Recommended 16GB
Comfortable
Gemma 4 12B (BF16)
Large language models (LLM)
Full precision
Minimum 26GB / Recommended 32GB
Comfortable
Gemma 4 E4B (Q8)
Large language models (LLM)
Near full quality
Minimum 10GB / Recommended 12GB
Comfortable
Gemma 4 E4B (BF16)
Large language models (LLM)
Full precision
Minimum 17GB / Recommended 20GB
Comfortable
Mistral Small 3.2 24B (Q3)
Large language models (LLM)
For running on 12-16GB cards
Minimum 13GB / Recommended 16GB
Comfortable
Mistral Small 3.2 24B (Q8)
Large language models (LLM)
Near full quality
Minimum 27GB / Recommended 32GB
Comfortable
GLM-4.7-Flash (Q3)
Large language models (LLM)
For running on 16-20GB cards
Minimum 17GB / Recommended 20GB
Comfortable
GLM-4.7-Flash (Q8)
Large language models (LLM)
Near full quality
Minimum 34GB / Recommended 48GB
Comfortable
Qwen3 8B (Q8)
Large language models (LLM)
Near full quality
Minimum 11GB / Recommended 12GB
Comfortable
Qwen3 8B (BF16)
Large language models (LLM)
Full precision
Minimum 18GB / Recommended 20GB
Comfortable
Llama 3.3 70B (Q3)
Large language models (LLM)
Better quality than Q2, for 48GB
Minimum 36GB / Recommended 48GB
Comfortable
Llama 3.3 70B (Q5)
Large language models (LLM)
Balanced quantization
Minimum 52GB / Recommended 64GB
