whatmodel.sh · find the best AI models you can run

Best AI models to run locally on a 8GB PC without a graphics card

A leaderboard of the best AI models you can run locally on a computer with 8GB of RAM and no graphics card. MiniCPM5 2B is #1 right now, followed by Spark-X2.5 1.7B and LFM2.5 2.6B. Models are ranked by one score that combines 575 published benchmarks. Each model shows the download that fits and the command to run it.

Filter by your hardware and budget → ~3GB for models (CPU, slow) · local only · reset

ram[2][4][8][16][24][32][48][64][96][128][192][256]GB

gpu[apple silicon][graphics card][unified memory][none]

price[local][$0.02][$0.1][$0.5][$2][$10][$50][any]max per M

Filter by task

task[all tasks][agent tasks][coding][chat & writing][math][vision & documents][web search]

New #1s for your hardware:

orRSS

detecting your OS…

$ terminal

Score vs release date

     │                                                                 EXAONE Deep 2.4B       MiniCPM5 2B ●
  50 ┤ · · · · · · · · · · · · · · · · · ·DeepSeek-Coder 1.3B Instr… ●─────────●●─∘●∘∘∘●∘──∘─∘∘───∘─∘───●●┘· ·
     │                                                  ●───────────●∘  ∘∘∘∘ ∘ ∘∘ ∘ ∘∘ ∘∘∘ ∘∘ ∘
   0 ┤ · · ·●──●─●─∘────────────────∘──────────────∘─∘──┘· ·∘· · · · · ·∘· · · · · · · · · · · · ··· · · · · ·
     │   ●──┘                                               ∘
     └───────────────┬────────────────────────┬─────────────────────────┬─────────────────────────┬───────────
                  2020-01                  2022-01                   2024-01                   2026-01

#	model	score↓	where↓	RAM↓	speed↓	context↓	get it	sources

#	model	score↓	where↓	RAM↓	speed↓	context↓	get it	sources
196	MiniCPM5 2B · 4-bit OpenBMB	63±1†	local (slow)	~2.5GB		131k	llama.cpp ›
331	Spark-X2.5 1.7B · 4-bit iFLYTEK (SparkLLM)	55±1†	local (slow)	~1.9GB	~166 tok/s	1M	llama.cpp ›
390	LFM2.5 2.6B · 4-bit Liquid AI	53±1	local (slow)	~2.4GB		131k	llama.cpp ›
447	EXAONE Deep 2.4B · 4-bit LG AI Research	50±2†	local (slow)	~3.2GB		33k	Ollama ›
452	LFM2.5 1.2B Thinking · 4-bit Liquid AI	50±2	local (slow)	~1.3GB		33k	llama.cpp ›
464	Qwen3-VL 2B Thinking · 4-bit Alibaba	49±1	local (slow)	~2.7GB		256k	llama.cpp ›
474	Qwen2.5 Coder 3B Instruct · 4-bit Alibaba	49±1	local (slow)	~2.8GB	~9.5 tok/s	33k	Ollama ›
508	Qwen3.5 2B · 4-bit Alibaba	47±1	local (slow)	~2.2GB	~174 tok/s	262k	llama.cpp ›	+7
526	Qwen2.5-Math 1.5B Instruct · 4-bit Alibaba	47±2†	local (slow)	~1.9GB		4k	llama.cpp ›
527	EXAONE 4.0 1.2B · 4-bit LG AI Research	47±1†	local (slow)	~2.8GB		66k	llama.cpp ›
546	Qwen2-Math 1.5B Instruct · 4-bit Alibaba	46±2†	local (slow)	~2.5GB		4k	Ollama ›
584	Qwen2.5-VL 3B Instruct · 4-bit Alibaba	45±1	local (slow)	~2.8GB			llama.cpp ›
585	Qwen3-VL 2B Instruct · 4-bit Alibaba	45±1	local (slow)	~2.7GB		256k	llama.cpp ›
586	LFM2.5 1.2B Instruct · 4-bit Liquid AI	45±2	local (slow)	~1.3GB		33k	llama.cpp ›
587	Qwen3 1.7B · 4-bit Alibaba	45±1	local (slow)	~2.9GB	~96 tok/s	33k	llama.cpp ›	+11
594	Qwen2.5 3B Instruct · 4-bit Alibaba	44±1	local (slow)	~3GB	~61 tok/s	33k	llama.cpp ›	+5
595	LFM2 2.6B · 4-bit Liquid AI	44±1	local (slow)	~2.2GB		33k	llama.cpp ›
604	DeepSeek-Coder 1.3B Instruct · 4-bit DeepSeek	44±1	local (slow)	~2.9GB		16k	Ollama ›
607	Qwen2.5 Coder 1.5B Instruct · 4-bit Alibaba	44±1	local (slow)	~1.9GB		33k	Ollama ›
638	Granite 4.0 1B · 4-bit IBM	42±1	local (slow)	~2.6GB		128k	llama.cpp ›
640	Granite 4.0 H 1B · 4-bit IBM	42±1	local (slow)	~2.4GB		131k	llama.cpp ›
653	Bielik 1.5B v3.0 Instruct · 4-bit SpeakLeash	41±3	local (slow)	~2.6GB			llama.cpp ›
655	Granite 3.3 2B Instruct · 4-bit IBM	41±1†	local (slow)	~3.1GB		128k	llama.cpp ›
659	LFM2 1.2B · 4-bit Liquid AI	41±1	local (slow)	~1.3GB		33k	llama.cpp ›
661	Granite 3.0 2B Instruct · 4-bit IBM	41±1	local (slow)	~3.2GB		4k	llama.cpp ›
692	Granite 3.1 2B Instruct · 4-bit IBM	39±1	local (slow)	~3.1GB		128k	llama.cpp ›
696	Granite 3.2 2B Instruct · 4-bit IBM	39±1	local (slow)	~3.1GB		131k	llama.cpp ›
701	Qwen2.5 Coder 0.5B Instruct · 4-bit Alibaba	38±1	local (slow)	~1.1GB		33k	Ollama ›
705	Qwen2-VL 2B Instruct · 4-bit Alibaba	38±1	local (slow)	~2.6GB		33k	llama.cpp ›
706	Qwen2.5 1.5B Instruct · 4-bit Alibaba	38±1	local (slow)	~2GB	~98 tok/s	33k	llama.cpp ›	+6
711	LFM2 700M · 4-bit Liquid AI	38±1	local (slow)	~1GB		33k	llama.cpp ›
712	Qwen3.5 0.8B · 4-bit Alibaba	38±1	local (slow)	~1.4GB	~319 tok/s	262k	llama.cpp ›	+2
714	EXAONE 3.5 2.4B Instruct · 4-bit LG AI Research	38±2	local (slow)	~3.2GB		33k	llama.cpp ›
725	DeepSeek R1 Distill Qwen 1.5B · 4-bit DeepSeek	37±2	local (slow)	~2GB		131k	llama.cpp ›	+1
733	Qwen3 0.6B · 4-bit Alibaba	36±1	local (slow)	~2.1GB	~61 tok/s	33k	llama.cpp ›	+5
746	OLMo 2 1B Instruct (0425) · 4-bit Allen Institute for AI	35±1	local (slow)	~2.4GB		4k	llama.cpp ›
751	DeepSeek-Coder 1.3B Base · 4-bit DeepSeek	34±1	local (slow)	~2.4GB		16k	Ollama ›
757	CodeGemma 2B (v1.0) · 4-bit Google	34±1	local (slow)	~3.2GB			llama.cpp ›
758	Qwen2 1.5B Instruct · 4-bit Alibaba	34±1	local (slow)	~1.9GB		32k	llama.cpp ›	+1
759	Falcon-H1 1.5B Instruct · 4-bit TII	33±4	local (slow)	~2.4GB		131k	llama.cpp ›
761	Gemma 3 1B · 4-bit Google	33±1	local (slow)	~1.8GB	~180 tok/s	32k	llama.cpp ›	+7
762	SmolLM2 1.7B Instruct · 4-bit Hugging Face	33±1	local (slow)	~3.1GB		8k	llama.cpp ›
763	LFM2 350M · 4-bit Liquid AI	33±1	local (slow)	~1.7GB		33k	llama.cpp ›
769	Llama 3.2 1B Instruct · 4-bit Meta	33±1	local (slow)	~1.7GB	~251 tok/s	131k	llama.cpp ›	+13
770	Phi-1.5 (1.3B) · 4-bit Microsoft	33±1	local (slow)	~2.4GB		2k	llama.cpp ›
779	Granite 3.1 1B-A400M Instruct · 4-bit IBM	31±2†	local (slow)	~2.3GB	~20 tok/s	131k	llama.cpp ›
781	Gemma 2B (base) · 4-bit Google	31±1	local (slow)	~3.2GB		8k	llama.cpp ›	+3
782	Gemma 1.1 2B IT · 4-bit Google	31±1	local (slow)	~3.2GB		8k	llama.cpp ›	+1
790	Qwen2.5 0.5B Instruct · 4-bit Alibaba	30±2	local (slow)	~1.2GB		33k	llama.cpp ›	+1
797	Granite 3.0 1B-A400M Instruct · 4-bit IBM	30±1	local (slow)	~2.3GB		4k	llama.cpp ›
801	EuroLLM 1.7B Instruct · 4-bit UTTER (EuroLLM)	29±3	local (slow)	~2.6GB		4k	llama.cpp ›
802	Granite 4.0 H 350M · 4-bit IBM	29±1†	local (slow)	~1.7GB		33k	llama.cpp ›
804	Gemma 2B IT · 4-bit Google	29±1	local (slow)	~3.1GB		8k	llama.cpp ›	+3
806	Granite 4.0 350M · 4-bit IBM	28±1	local (slow)	~1.7GB		32k	llama.cpp ›
810	Qwen2 0.5B Instruct · 4-bit Alibaba	28±2	local (slow)	~1.1GB		32k	llama.cpp ›
816	Falcon3 1B Instruct · 4-bit TII	26±4	local (slow)	~2.7GB		8k	llama.cpp ›
817	Qwen1.5 1.8B Chat · 4-bit Alibaba	26±2	local (slow)	~2.7GB		33k	Ollama ›
823	Flan-T5 Base · 4-bit Google	25±1	local (slow)	~2.5GB			how to run ›	+3
834	Qwen1.5 0.5B Chat · 4-bit Alibaba	23±3	local (slow)	~1.8GB		33k	Ollama ›
837	Phi-1 (1.3B) · 4-bit Microsoft	22±4	local (slow)	~2.4GB		2k	llama.cpp ›
839	SmolLM2 360M Instruct · 8-bit Hugging Face	22±3	local (slow)	~1.2GB		8k	llama.cpp ›
841	OLMo 1B · 4-bit Allen Institute for AI	21±2†	local (slow)	~2.2GB		2k	llama.cpp ›
851	Gemma 3 270M IT · 8-bit Google	17±4	local (slow)	~1.2GB		32k	llama.cpp ›
852	Flan-T5 Small · 4-bit Google	16±3	local (slow)	~1.8GB			how to run ›
857	T5 Base · 4-bit Google	14±3	local (slow)	~2.4GB		1k	how to run ›
858	Pythia 1.4B · 4-bit EleutherAI	14±1	local (slow)	~2.4GB		2k	llama.cpp ›
859	Pythia 1B · 4-bit EleutherAI	12±1	local (slow)	~2.2GB		2k	llama.cpp ›
863	T5 Small · 4-bit Google	12±3	local (slow)	~1.8GB		1k	how to run ›
864	BLOOM 1.7B · 4-bit BigScience	11±1	local (slow)	~3.1GB		2k	llama.cpp ›	+1
865	Zyte-1B · 4-bit venkycs (aihub-app)	11±2	local (slow)	~2.2GB			llama.cpp ›
868	Pythia 410M · 4-bit EleutherAI	8±2	local (slow)	~1.8GB		2k	llama.cpp ›
870	BLOOM 1.1B · 4-bit BigScience	7±1	local (slow)	~2.6GB		2k	llama.cpp ›
871	GPT-2 XL (1.5B) · 4-bit OpenAI	7±1	local (slow)	~2.8GB	~75 tok/s	1k	llama.cpp ›	+9
874	OPT-350M · 4-bit Meta	1±1	local (slow)	~2.2GB		2k	how to run ›	+1
877	BLOOM 560M · 4-bit BigScience	0±1	local (slow)	~2.1GB		2k	llama.cpp ›	+1
878	Pythia 160M · 4-bit EleutherAI	-2±2	local (slow)	~1.6GB		2k	llama.cpp ›
879	GPT-2 Large (774M) · 4-bit OpenAI	-3±1	local (slow)	~2GB	~137 tok/s	1k	llama.cpp ›	+3
880	OPT-125M · 4-bit Meta	-3±1	local (slow)	~1.8GB		2k	how to run ›	+1
881	GPT-Neo 125M · 4-bit EleutherAI	-3±2	local (slow)	~2.1GB		2k	how to run ›
882	GPT-2 Medium (355M) · 4-bit OpenAI	-5±1	local (slow)	~1.8GB		1k	llama.cpp ›	+3
883	Pythia 70M · 4-bit EleutherAI	-13±2	local (slow)	~1.6GB		2k	llama.cpp ›
884	GPT-2 Small (124M) · 4-bit OpenAI	-13±2	local (slow)	~1.6GB		1k	llama.cpp ›	+3
—	LFM3 1.7B · q4_k_m Liquid AI	not yet calibrated	local (slow)	~2.6GB		262k	llama.cpp ›

+ 836 more don't fit your machine or budget · show them

+ 16 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them

919 of 1227 catalogue models listed here · sources and licences

* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory