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

Best AI models for math to run locally on a 16GB graphics card

A leaderboard of the best AI models for math you can run locally on a 16GB graphics card (with 32GB of system RAM). Qwen3-Coder-Next (80B-A3B) is #1 right now, followed by EXAONE 4.5 33B and Granite 4.2 30B. Models are ranked on math tests, from competition problems like AIME to research problems like FrontierMath. Each model shows the download that fits and the command to run it.

264 models · scores combine 59 benchmarks · updated 2026-10-11 02:20 UTChow it works · find a leaderboard · benchmark your machine

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

Scores use reviewed math evaluations only. Compare scores within this task. Models without task measurements are excluded.

Models for 32GB RAM · 16GB graphics card → ~16GB fast · local only · reset

+ more

card[RTX 4060 Ti 16GB][RTX 4070 Ti Super][RTX 4080][RTX 4080 Super][RTX 5060 Ti 16GB][RTX 5070 Ti][RTX 5080][RTX 3080 Laptop][RTX 3080 Ti Laptop][RTX 4090 Laptop][RTX 5080 Laptop][RTX A4000][RX 7600 XT][RX 7800 XT][RX 7900 GRE][RX 9060 XT][RX 9070][RX 9070 XT]

Filter the model list while keeping task scores. Models without a shared score scale are not yet calibrated; select a set to see its own scores.

Detect your exact chip or graphics card:

curl -G whatmodel.sh/ -d max=0 --data-urlencode "mac=$(sysctl -n hw.memsize machdep.cpu.brand_string)"Mac

curl -G whatmodel.sh/ -d max=0 --data-urlencode "ram=$(awk '/MemT/{print $2}' /proc/meminfo)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>/dev/null)"Linux

curl.exe -G whatmodel.sh/ -d max=0 --data-urlencode "ram=$((gcim Win32_ComputerSystem).TotalPhysicalMemory)" --data-urlencode "nv=$(nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null)"Windows

$ terminal

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

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

vram[1][2][4][8][12][16][24][32][48][64][80][96][128][160]GB

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

detecting your OS…

Score vs memory to run

     │     MiniCPM5 2B          ●─Spark-X2.5 4B─────────────────────────────────────∘─●──∘──●───●       ∘
  60 ┤ · ●────────────●─────────∘· · · · · · ·∘· · ∘ · ∘ · ∘ · ∘ ·∘· · · ·∘· · ·∘·∘·∘··∘ ∘∘·∘∘ ·∘· ∘ · ∘∘· · ·
  40 ┤ · ∘ · · · · · ·∘· · · · ·∘· · · ∘ · · ·∘· · ∘ · ∘ · ∘ · ∘ ·∘· ∘ · · · · ·∘· ·∘· ∘ ∘∘·∘· ·∘∘∘∘ ∘∘∘ ∘∘· ·
  20 ┤ · ∘ · · · · · ·∘· · · · ·∘· · · · · · ·∘· · ∘ · · · ∘ · ∘ · · ∘ · · · ·∘· · · ··· ·∘·∘· ·∘·∘·∘∘ ·∘· · ·
     │   ∘            ∘                ∘      ∘
     └───┬─────────────────────────────┬───────────────────────┬──────────────────────┬───────────────────────
        2GB                           5GB                    10GB                   20GB

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

#	model	task score↓	where↓	RAM↓	context↓	get it	sources
66	Qwen3-Coder-Next (80B-A3B) · 2-bit Alibaba	66±5*	local (slow)	~27GB	262k	llama.cpp ›
71	EXAONE 4.5 33B · 4-bit LG AI Research	65±4*†	local (slow)	~24GB	262k	llama.cpp ›
74	Granite 4.2 30B · 4-bit IBM	65±3*†	local (slow)	~20GB	131k	llama.cpp ›
84	Qwen3.6 27B · 4-bit Alibaba	64±1	local (slow)	~20GB	262k	Ollama ›
92	Spark-X2.5 4B · 16-bit iFLYTEK (SparkLLM)	64±2†	local	~10GB	1M	Ollama ›
98	GLM-4.5-Air · 2-bit Z.ai	63±2	local (slow)	~35GB	131k	llama.cpp ›
99	Qwen3.5 27B · 4-bit Alibaba	63±1	local (slow)	~19GB	262k	llama.cpp ›
101	Nemotron Cascade 2 30B-A3B · 4-bit NVIDIA	63±2†	local (slow)	~22GB	256k	Ollama ›
104	Gemma 4 26B-A4B · 4-bit QAT Google	63±2	local (slow)	~18GB	262k	Ollama ›
105	Qwen3-Next 80B-A3B Instruct · 2-bit Alibaba	63±3†	local (slow)	~27GB	131k	llama.cpp ›
112	Qwen3-VL 32B Thinking · 4-bit Alibaba	62±2	local (slow)	~22GB	256k	llama.cpp ›
117	Nemotron-Labs-3 Puzzle 75B-A9B · 2-bit NVIDIA	62±3*†	local (slow)	~35GB		llama.cpp ›
120	Olmo 3.1 32B Think · 4-bit Ai2	62±2†	local (slow)	~22GB	66k	llama.cpp ›
123	Gemma 4 31B · 4-bit Google	61±2	local (slow)	~21GB	262k	llama.cpp ›
128	Nemotron-Cascade-14B-Thinking · 4-bit NVIDIA	61±2*†	local	~11GB		llama.cpp ›
130	Granite 4.2 8B · 8-bit IBM	61±3*†	local	~11GB	131k	llama.cpp ›
131	Falcon-H1R 7B · 8-bit TII	61±2†	local	~12GB	262k	llama.cpp ›
132	Qwen3-Next 80B-A3B Thinking · 2-bit Alibaba	61±2†	local (slow)	~27GB	131k	llama.cpp ›
133	Qwen3-VL 30B-A3B Thinking · 4-bit Alibaba	61±3*†	local (slow)	~21GB	256k	llama.cpp ›
134	Qwen3.5 35B-A3B · 4-bit Alibaba	61±1	local (slow)	~24GB	262k	llama.cpp ›
135	Qwen3.6 35B-A3B · 4-bit Alibaba	61±1	local (slow)	~23GB	262k	llama.cpp ›
139	Nemotron-Cascade-8B · 8-bit NVIDIA	60±2*†	local	~11GB		llama.cpp ›
143	Olmo 3 32B Think · 4-bit Ai2	60±1†	local (slow)	~22GB	66k	llama.cpp ›
146	gpt-oss-20b · MXFP4 OpenAI	60±1	local	~15GB	131k	Ollama ›
147	MiniCPM5 2B · 8-bit OpenBMB	60±2†	local	~4.3GB	131k	llama.cpp ›
149	Seed-OSS 36B Instruct · 4-bit ByteDance Seed	60±2	local (slow)	~25GB	524k	llama.cpp ›
150	Qwen3 30B-A3B Instruct 2507 · 4-bit Alibaba	59±1	local (slow)	~21GB	262k	llama.cpp ›
153	Nemotron 3 Nano 30B-A3B · 4-bit NVIDIA	59±2†	local (slow)	~22GB	262k	Ollama ›
155	Qwen3.5 9B · 8-bit Alibaba	59±1	local	~12GB	262k	llama.cpp ›
156	Qwen3 14B · 4-bit Alibaba	59±2*	local	~11GB	33k	Ollama ›
157	QwQ-32B · 4-bit Alibaba	58±2	local (slow)	~22GB	131k	llama.cpp ›
158	Qwen3-VL 8B Thinking · 8-bit Alibaba	58±2	local	~12GB	256k	Ollama ›
160	Qwen3.5 4B · 8-bit Alibaba	58±1	local	~6.2GB	262k	llama.cpp ›
162	EXAONE 4.0 32B · 4-bit LG AI Research	58±2†	local (slow)	~22GB	131k	llama.cpp ›
163	Granite 4.2 3B · 8-bit IBM	58±1	local	~5.6GB	131k	llama.cpp ›
164	DeepSeek R1 0528 Qwen3 8B · 8-bit DeepSeek	58±1	local	~12GB	128k	Ollama ›
165	Qwen3-VL 4B Thinking · 8-bit Alibaba	57±2*†	local	~6.9GB	256k	Ollama ›
166	Qwen3 30B-A3B Thinking 2507 · 4-bit Alibaba	57±1	local (slow)	~21GB	262k	llama.cpp ›
169	Olmo 3 7B Think · 8-bit Ai2	57±1†	local	~11GB	66k	Ollama ›
171	Phi-4-reasoning-plus 14B · 4-bit Microsoft	57±1	local	~13GB		Ollama ›
172	Qwen3-VL 32B Instruct · 4-bit Alibaba	57±2*†	local (slow)	~22GB	256k	llama.cpp ›
173	GLM-4.7-Flash · 4-bit Z.ai	57±2*	local (slow)	~21GB	200k	llama.cpp ›
174	Gemma 4 12B · 8-bit Google	57±2	local	~15GB	262k	llama.cpp ›
177	Nemotron Nano 9B v2 · 8-bit NVIDIA	57±1	local	~13GB	131k	llama.cpp ›
179	Qwen3-VL 30B-A3B Instruct · 4-bit Alibaba	56±2*†	local (slow)	~21GB	256k	llama.cpp ›
181	DeepSeek R1 Distill Llama 70B · 2-bit DeepSeek	56±1	local (slow)	~30GB	131k	llama.cpp ›
184	OpenThinker3 7B · 8-bit Open Thoughts	56±1	local	~10GB	33k	llama.cpp ›
185	OpenThinker2 32B · 4-bit Open Thoughts	56±2†	local (slow)	~22GB	33k	llama.cpp ›
186	Qwen3 32B · 4-bit Alibaba	56±1	local (slow)	~22GB	131k	llama.cpp ›
187	Qwen3 8B · 8-bit Alibaba	55±1	local	~12GB	33k	Ollama ›
188	DeepSeek R1 Distill Qwen 32B · 4-bit DeepSeek	55±1	local (slow)	~22GB	131k	llama.cpp ›
189	DeepSeek R1 Distill Qwen 14B · 4-bit DeepSeek	55±1	local	~11GB	131k	Ollama ›
190	Phi-4-reasoning 14B · 4-bit Microsoft	55±2†	local	~11GB	33k	Ollama ›
196	Qwen3 4B Instruct 2507 · 8-bit Alibaba	54±1	local	~7.1GB	262k	Ollama ›
197	Phi-4-mini-reasoning 3.8B · 8-bit Microsoft	54±2*†	local	~5.8GB	128k	llama.cpp ›
198	Magistral Small 1.2 · 4-bit Mistral	54±1	local (slow)	~17GB	131k	llama.cpp ›
201	OpenThinker 32B · 4-bit Open Thoughts	53±1†	local (slow)	~22GB	33k	llama.cpp ›
202	Qwen3 4B Thinking 2507 · 8-bit Alibaba	53±2	local	~7.1GB	262k	Ollama ›
206	QED-Nano · 16-bit LM Provers	52±2	local	~9.9GB	262k	how to run ›
207	Qwen3-VL 8B Instruct · 8-bit Alibaba	52±2	local	~12GB	256k	Ollama ›
210	DeepSeek R1 Distill Qwen 7B · 8-bit DeepSeek	52±1	local	~12GB	131k	Ollama ›	+1
211	EXAONE Deep 32B · 4-bit LG AI Research	52±2†	local (slow)	~22GB	33k	Ollama ›
212	Nemotron 3 Nano 4B · 8-bit NVIDIA	51±1	local	~7GB	262k	Ollama ›
213	s1.1 32B · 4-bit Simple Scaling	51±1	local (slow)	~22GB	33k	llama.cpp ›
214	Spark-X2.5 1.7B · 16-bit iFLYTEK (SparkLLM)	51±2†	local	~5.1GB	1M	Ollama ›
217	Qwen3 30B-A3B (Apr 2025, hybrid) · 4-bit Alibaba	50±1	local (slow)	~21GB	33k	llama.cpp ›
219	EXAONE Deep 7.8B · 4-bit LG AI Research	50±2†	local	~6.5GB	33k	Ollama ›
220	OpenThinker2 7B · 4-bit Open Thoughts	50±2†	local	~6.4GB	33k	llama.cpp ›
222	Gemma 3 27B · 4-bit Google	50±1	local (slow)	~19GB	131k	llama.cpp ›
223	Ministral 3 14B · 4-bit Mistral	50±3*†	local	~11GB	262k	Ollama ›
224	Magistral Small 1.0 (2506) · 4-bit Mistral	50±1	local (slow)	~17GB	128k	llama.cpp ›
225	Mistral Small 3.2 24B · 4-bit Mistral	50±2*	local (slow)	~17GB	128k	llama.cpp ›
228	Olmo 3 7B Instruct · 8-bit Ai2	49±1†	local	~11GB	66k	Ollama ›
231	Gemma 3 12B · 8-bit Google	49±1	local	~15GB	131k	llama.cpp ›
232	DeepSeek R1 Distill Llama 8B · 8-bit DeepSeek	49±1	local	~12GB	131k	Ollama ›
233	Magistral Small 1.1 (2507) · 4-bit Mistral	49±2*†	local (slow)	~17GB	40k	llama.cpp ›
235	Qwen3-VL 4B Instruct · 8-bit Alibaba	49±2*†	local	~6.9GB	256k	Ollama ›
238	Command A Reasoning (08-2025) · 2-bit Cohere	49±4*†	local (slow)	~34GB		llama.cpp ›
239	LFM2.5 1.2B Thinking · 8-bit Liquid AI	49±3†	local	~3.5GB	33k	Ollama ›
240	Ministral 3 8B · 8-bit Mistral	48±2	local	~13GB	262k	Ollama ›
242	LFM2.5 8B-A1B · 8-bit Liquid AI	48±1	local	~13GB	128k	Ollama ›
244	Qwen3 4B · 8-bit Alibaba	48±1	local	~7.1GB	33k	Ollama ›
245	LFM2 24B-A2B · 4-bit Liquid AI	48±2	local (slow)	~17GB	33k	llama.cpp ›
249	Phi-4 14B · 4-bit Microsoft	48±1	local	~11GB	16k	Ollama ›
251	Qwen3-VL 2B Thinking · 8-bit Alibaba	48±2*†	local	~4.2GB	256k	Ollama ›
254	Qwen2.5 72B Instruct · 2-bit Alibaba	47±1	local (slow)	~30GB	33k	llama.cpp ›
255	Llama 4 Scout (109B-A17B) · 2-bit Meta	47±1*	local (slow)	~36GB	328k	llama.cpp ›
257	Qwen2-Math 72B Instruct · 2-bit Alibaba	47±2†	local (slow)	~32GB	4k	llama.cpp ›
258	Granite 4.0 H Small (32B-A9B) · 4-bit IBM	47±1	local (slow)	~22GB	131k	llama.cpp ›
261	Tulu 2 70B · 2-bit Allen Institute for AI	46±5*†	local (slow)	~28GB		llama.cpp ›
262	OpenThinker 7B · 8-bit Open Thoughts	46±2†	local	~10GB	33k	llama.cpp ›
264	EXAONE Deep 2.4B · 4-bit LG AI Research	46±2†	local	~3.2GB	33k	llama.cpp ›
265	Qwen2.5-Math 7B Instruct · 4-bit Alibaba	46±2†	local	~6.4GB	4k	llama.cpp ›
266	Qwen2.5 32B Instruct · 4-bit Alibaba	46±1	local (slow)	~22GB	131k	llama.cpp ›
267	Llama 3.1 Nemotron 51B Instruct · 4-bit NVIDIA	46±5*†	local (slow)	~34GB		llama.cpp ›
268	LFM2.5 2.6B · 8-bit Liquid AI	46±1	local	~4.5GB	131k	llama.cpp ›
271	Tulu 2 DPO 70B · 2-bit Ai2	45±5*†	local (slow)	~32GB		llama.cpp ›
273	Qwen2.5 14B Instruct · 4-bit Alibaba	45±2	local	~11GB	131k	Ollama ›
274	Granite 4.0 H Tiny (7B-A1B) · 4-bit IBM	45±2†	local	~5.9GB	131k	llama.cpp ›
275	DeepSeek R1 Distill Qwen 1.5B · 8-bit DeepSeek	45±1	local	~4.3GB	131k	Ollama ›
282	Qwen2.5-VL 7B Instruct · 8-bit Alibaba	45±3*†	local	~11GB		Ollama ›
283	Command A (03-2025) · 2-bit Cohere	44±3*†	local (slow)	~37GB	256k	llama.cpp ›
284	Gemma 4 E4B · 8-bit Google	44±1	local	~9.9GB	131k	llama.cpp ›
286	Qwen2-VL 72B Instruct · 2-bit Alibaba	44±3	local (slow)	~34GB		llama.cpp ›
290	Mixtral 8x7B v0.1 (base) · 4-bit Mistral AI	44±2*†	local (slow)	~29GB	32k	llama.cpp ›
291	EXAONE 4.0 1.2B · 8-bit LG AI Research	44±2*†	local	~3.6GB	66k	llama.cpp ›
292	Qwen2.5 7B Instruct · 8-bit Alibaba	44±1	local	~12GB	131k	Ollama ›
293	SmolLM3 3B · 8-bit Hugging Face	44±2	local	~5.9GB	66k	llama.cpp ›
294	Gemma 3 4B · 8-bit Google	43±2	local	~7.4GB	131k	Ollama ›
295	Mistral Small 3.1 24B (2503) · 4-bit Mistral	43±1	local (slow)	~17GB	128k	llama.cpp ›
297	Granite 3.3 8B Instruct · 8-bit IBM	43±1	local	~11GB	128k	llama.cpp ›
302	InternVL2-4B · 16-bit Shanghai AI Laboratory (OpenGVLab)	43±5*	local	~10GB		how to run ›
304	Xwin-LM 70B v0.1 · 2-bit Xwin-LM	43±5*	local (slow)	~28GB		llama.cpp ›
305	Qwen3 1.7B · 8-bit Alibaba	43±1	local	~4.6GB	33k	Ollama ›	+1
306	Mistral Small 3 24B (2501) · 4-bit Mistral	43±1*	local (slow)	~17GB	32k	llama.cpp ›
307	Granite 4.0 H Micro 3B · 4-bit IBM	43±2†	local	~3.5GB	131k	llama.cpp ›
308	Llama 3.1 Tulu 3 70B DPO · 2-bit Ai2	43±1	local (slow)	~22GB	131k	llama.cpp ›
309	Llama 3.3 70B Instruct · 2-bit Meta	42±1	local (slow)	~24GB	128k	llama.cpp ›
310	Ministral 3 3B · 8-bit Mistral	42±1	local	~6.9GB	262k	Ollama ›
312	Qwen2-Math 7B Instruct · 8-bit Alibaba	42±2†	local	~10GB	4k	llama.cpp ›
313	Granite 4.0 Micro 3B · 8-bit IBM	42±2	local	~5.3GB	131k	llama.cpp ›
315	Granite 8B Code Base (4K) · 4-bit IBM	42±2*†	local	~6.6GB		llama.cpp ›
318	Qwen2 72B Instruct · 2-bit Alibaba	41±1	local (slow)	~33GB	131k	llama.cpp ›
319	Llama 3.1 70B Instruct · 2-bit Meta	41±1	local (slow)	~27GB	128k	llama.cpp ›
322	Gemma 4 E2B · 8-bit Google	41±1	local	~6.8GB	131k	llama.cpp ›
323	Qwen2.5-Math 1.5B Instruct · 4-bit Alibaba	41±2†	local	~2.5GB	4k	llama.cpp ›
325	Phi-4-mini 3.8B · 8-bit Microsoft	41±2†	local	~6.9GB	128k	Ollama ›
327	Phi-2 (2.7B) · 4-bit Microsoft	40±3*	local	~7.3GB	2k	how to run ›
328	Molmo 2 8B · 8-bit Ai2	40±7*†	local	~11GB		llama.cpp ›
329	Pixtral 12B (2409) · 8-bit Mistral	40±3*†	local	~15GB		llama.cpp ›
330	StarCoder2 15B · 4-bit BigCode	40±2*†	local	~12GB	16k	Ollama ›
331	LFM2 2.6B · 4-bit Liquid AI	40±3*†	local	~3.1GB	33k	llama.cpp ›
332	Qwen3.5 2B · 8-bit Alibaba	40±1	local	~4.9GB	262k	Ollama ›
334	Qwen2.5-VL 3B Instruct · 8-bit Alibaba	40±3*†	local	~6.3GB		Ollama ›
336	Qwen3-VL 2B Instruct · 8-bit Alibaba	40±2*†	local	~4.2GB	256k	Ollama ›
337	Gemma 7B (base) · 8-bit Google	40±2	local	~11GB	8k	Ollama ›
338	InternLM2.5 7B Chat · 8-bit Shanghai AI Laboratory (InternLM)	39±3*†	local	~12GB		Ollama ›
339	Qwen2-Math 1.5B Instruct · 8-bit Alibaba	39±2†	local	~3.2GB	4k	llama.cpp ›
340	CodeGemma 7B (v1.0) · 8-bit Google	39±2*	local	~11GB		Ollama ›
342	LFM2 1.2B · 4-bit Liquid AI	39±3*†	local	~2.3GB	33k	llama.cpp ›
343	WizardLM 70B V1.0 · 2-bit Microsoft	38±4*	local (slow)	~32GB		llama.cpp ›
345	Llama 3.2 11B Vision Instruct · 4-bit Meta	38±3*	local	~8.8GB	128k	Ollama ›
347	LFM2.5 1.2B Instruct · 4-bit Liquid AI	38±2†	local	~2.3GB	33k	llama.cpp ›
348	Gemma 2 27B IT · 4-bit Google	38±1	local (slow)	~19GB	8k	llama.cpp ›
350	Llama 2 70B (base) · 2-bit Meta	38±2	local (slow)	~32GB	4k	llama.cpp ›
351	Phi-3-vision-128k-instruct · 16-bit Microsoft	38±6*†	local	~10GB		how to run ›
352	Mathstral 7B v0.1 · 8-bit Mistral AI	38±3*†	local	~11GB	32k	Ollama ›
353	Granite 3.3 2B Instruct · 8-bit IBM	38±2†	local	~4.3GB	128k	llama.cpp ›
355	Phi-3.5-vision-instruct · 16-bit Microsoft	38±7*†	local	~10GB		how to run ›
356	DeepSeek-Coder 33B Base · 4-bit DeepSeek	38±2*	local (slow)	~22GB	16k	llama.cpp ›
358	Tulu 2 DPO 13B · 4-bit Allen Institute for AI	37±5*†	local	~9.8GB		llama.cpp ›
359	Llama 3 70B Instruct · 2-bit Meta	37±1	local (slow)	~24GB	8k	llama.cpp ›
360	Jamba 1.5 Mini · 4-bit AI21 Labs	37±4*†	local (slow)	~34GB		llama.cpp ›
361	Yi 1.5 34B Chat · 4-bit 01.AI	37±1	local (slow)	~23GB	4k	llama.cpp ›
362	Llama 3.1 Tulu 3 8B · 8-bit Ai2	37±1	local	~12GB	131k	Ollama ›
364	Granite 3.2 8B Instruct · 8-bit IBM	37±2†	local	~12GB	131k	Ollama ›
366	Granite 4.0 1B · 16-bit IBM	37±2	local	~6.1GB	128k	Ollama ›
367	Qwen2.5 3B Instruct · 8-bit Alibaba	37±1	local	~5.9GB	33k	Ollama ›
368	OLMo 2 32B Instruct (0325) · 4-bit Ai2	37±1†	local (slow)	~21GB		llama.cpp ›
369	Llama 3.1 8B Instruct · 8-bit Meta	37±1	local	~12GB	128k	Ollama ›
370	Granite 3.1 8B Instruct · 8-bit IBM	36±2†	local	~12GB	128k	Ollama ›
371	OLMo 2 13B Instruct (1124) · 4-bit Allen Institute for AI	36±2*†	local	~11GB	4k	Ollama ›
372	Gemma 2 9B IT · 8-bit Google	36±1	local	~14GB	8k	Ollama ›
373	Tulu 2 13B · 4-bit Allen Institute for AI	36±4*†	local	~9.8GB		llama.cpp ›
374	Granite 3.0 8B Instruct · 8-bit IBM	36±2*†	local	~12GB	4k	Ollama ›
377	Granite 4.0 H 1B · 8-bit IBM	36±2	local	~3.2GB	131k	Ollama ›
378	Athene-70B · 2-bit Nexusflow	35±1*	local (slow)	~29GB		llama.cpp ›
379	Qwen3 0.6B · 8-bit Alibaba	35±2	local	~3GB	33k	Ollama ›
380	Phi-3 Medium 128K Instruct · 4-bit Microsoft	35±1*	local	~11GB	128k	Ollama ›
381	Code Llama 34B (base) · 4-bit Meta	35±2*	local (slow)	~23GB	16k	Ollama ›
382	Qwen2 7B Instruct · 8-bit Alibaba	35±1	local	~12GB	131k	Ollama ›
384	Ministral 8B (2410) · 8-bit Mistral	34±1	local	~10GB	128k	llama.cpp ›
385	Code Tulu 2 13B · 16-bit Allen Institute for AI	34±5*†	local (slow)	~29GB		how to run ›
387	LFM2 700M · 4-bit Liquid AI	34±3*†	local	~2GB	33k	llama.cpp ›
390	Phi-3.5-mini-instruct · 8-bit Microsoft	34±1	local	~6.8GB	128k	Ollama ›
392	Qwen2-VL 7B Instruct · 8-bit Alibaba	34±2	local	~11GB		llama.cpp ›
393	Code Llama 34B Instruct · 4-bit Meta	33±5*	local (slow)	~23GB	16k	llama.cpp ›
394	Yi 1.5 9B Chat · 8-bit 01.AI	33±1*	local	~13GB	4k	Ollama ›
396	Granite 3.0 2B Instruct · 8-bit IBM	33±2*†	local	~5.3GB	4k	Ollama ›
397	OLMo 2 7B Instruct (1124) · 8-bit Allen Institute for AI	33±2†	local	~11GB	4k	Ollama ›
398	Qwen3.5 0.8B · 8-bit Alibaba	33±4†	local	~3.1GB	262k	Ollama ›
399	Qwen2.5 1.5B Instruct · 8-bit Alibaba	33±2	local	~4GB	33k	Ollama ›
401	Mistral 7B v0.1 (base) · 8-bit Mistral AI	33±1	local	~9.6GB	8k	llama.cpp ›
402	Llama 3.2 3B Instruct · 8-bit Meta	32±1	local	~6.1GB	131k	Ollama ›
404	Mistral NeMo Instruct (2407) · 8-bit Mistral	32±1*	local	~15GB	128k	llama.cpp ›
406	Tulu 2 DPO 7B · 8-bit Allen Institute for AI	32±5*†	local	~9.1GB		llama.cpp ›
408	Tulu 2 7B · 8-bit Allen Institute for AI	32±5*†	local	~9.1GB		llama.cpp ›
409	Phi-3-mini-4k-instruct (June 2024 update) · 8-bit Microsoft	32±1*	local	~5.8GB		llama.cpp ›
411	Code Tulu 2 7B · 16-bit Allen Institute for AI	32±5*†	local (slow)	~16GB		how to run ›
412	Gemma 3 1B · 4-bit Google	31±1	local	~2.7GB	32k	Ollama ›
413	DeepSeek-Coder 6.7B Base · 8-bit DeepSeek	31±2*	local	~9.1GB	16k	Ollama ›
414	Granite 3.2 2B Instruct · 8-bit IBM	31±2†	local	~5.2GB	131k	Ollama ›
415	Granite 3.1 2B Instruct · 8-bit IBM	31±2†	local	~5.2GB	128k	Ollama ›
417	Stable Code 3B · 8-bit Stability AI	31±2*	local	~4.7GB	16k	Ollama ›
418	Mixtral 8x7B Instruct v0.1 · 4-bit Mistral	30±1*	local (slow)	~29GB	32k	llama.cpp ›
419	StarCoder2 7B · 8-bit BigCode	30±2	local	~9.5GB	16k	Ollama ›
420	Granite 3.1 3B-A800M Instruct · 8-bit IBM	30±4*†	local	~5.2GB	131k	llama.cpp ›
421	Zephyr 7B beta · 8-bit Hugging Face	30±5*	local	~11GB		Ollama ›
422	Granite 3.0 3B-A800M Instruct · 8-bit IBM	30±2*†	local	~5.3GB	4k	llama.cpp ›
423	OLMo 2 1B Instruct (0425) · 16-bit Allen Institute for AI	29±2*†	local	~5.8GB	4k	how to run ›
425	Code Llama 13B (base) · 4-bit Meta	29±2*	local	~9.8GB	16k	Ollama ›
426	Llama 3 8B Instruct · 8-bit Meta	29±1	local	~12GB	8k	Ollama ›
427	WizardLM 13B V1.2 · 4-bit Microsoft	29±4*	local	~9.8GB		llama.cpp ›
428	DeepSeek LLM 67B Chat · 2-bit DeepSeek	29±1*	local (slow)	~31GB	4k	llama.cpp ›
429	Apertus 8B Instruct (2509) · 8-bit Swiss AI	29±3*	local	~11GB	66k	llama.cpp ›
430	Code Llama 13B Instruct · 4-bit Meta	28±5*	local	~9.8GB	16k	Ollama ›
431	SmolLM2 1.7B Instruct · 8-bit Hugging Face	27±2	local	~4.2GB	8k	Ollama ›
432	Yi 34B Chat · 4-bit 01.AI	27±1*	local (slow)	~23GB	4k	llama.cpp ›
433	LFM2 350M · 4-bit Liquid AI	27±3*†	local	~1.7GB	33k	llama.cpp ›
434	Falcon 40B (base) · 4-bit TII	27±3*	local (slow)	~27GB	2k	Ollama ›
435	Gemma 2B (base) · 8-bit Google	27±4*†	local	~4.3GB	8k	Ollama ›
436	Llama 2 70B Chat · 2-bit Meta	26±1	local (slow)	~32GB	4k	llama.cpp ›
437	Llama 2 13B (base) · 4-bit Meta	26±2	local	~9.8GB	4k	llama.cpp ›
438	StarCoder2 3B · 8-bit BigCode	26±2*†	local	~4.9GB	16k	Ollama ›
439	Llama 3.2 1B Instruct · 4-bit Meta	25±1	local	~5.2GB	131k	Ollama ›
440	Code Llama 7B Instruct · 8-bit Meta	25±5*	local	~9.1GB	16k	Ollama ›
441	Llama-3-Instruct-8B-SimPO-v0.2 · 4-bit Princeton NLP	25±2*	local	~6.6GB		llama.cpp ›
442	Granite 3.1 1B-A400M Instruct · 8-bit IBM	25±4*†	local	~3GB	131k	Ollama ›
443	Mistral 7B Instruct v0.3 · 8-bit Mistral	25±1	local	~11GB	32k	Ollama ›
444	Qwen2 0.5B Instruct · 8-bit Alibaba	24±5*†	local	~2.6GB	32k	Ollama ›
445	Molmo 7B-D (0924) · 4-bit Ai2	24±3	local (slow)	~35GB		how to run ›
446	Gemma 2 2B IT · 8-bit Google	24±1	local	~5.3GB	8k	Ollama ›
447	Command R (03-2024) · 4-bit Cohere	23±1*	local (slow)	~24GB	128k	llama.cpp ›
448	Llama 2 7B Chat · 8-bit Meta	23±5*	local	~11GB	4k	Ollama ›
449	Qwen2-VL 2B Instruct · 4-bit Alibaba	23±3*†	local	~3.3GB	33k	llama.cpp ›
450	StarCoderBase 15.5B · 4-bit BigCode	23±2*†	local	~12GB		llama.cpp ›
451	Qwen2 1.5B Instruct · 8-bit Alibaba	23±1	local	~4GB	32k	Ollama ›
452	Granite 3.0 1B-A400M Instruct · 8-bit IBM	23±3*†	local	~3GB	4k	llama.cpp ›
453	Code Llama 7B (base) · 8-bit Meta	22±2	local	~9.1GB	16k	Ollama ›
455	Granite 4.0 H 350M · 8-bit IBM	20±2†	local	~1.9GB	33k	Ollama ›
456	Granite 4.0 350M · 16-bit IBM	19±2	local	~2.9GB	32k	Ollama ›
457	Llama 2 13B Chat · 4-bit Meta	19±3*	local	~10GB	4k	Ollama ›
458	Llama 2 7B (base) · 8-bit Meta	18±2	local	~9.1GB	4k	llama.cpp ›
460	DeepSeek-Coder 1.3B Base · 8-bit DeepSeek	17±3*	local	~3GB	16k	Ollama ›
461	StarCoderBase 7B · 4-bit BigCode	17±2	local	~6GB		Ollama ›
462	StarCoderBase 3B · 4-bit BigCode	13±3*†	local	~14GB		how to run ›
—	DeepSeek LLM 7B Chat · q4_0 DeepSeek	not yet calibrated	local	~5.7GB		Ollama ›
—	DeepSeek-V2 Lite Chat (16B-A2.4B) · Q4_K_M DeepSeek	not yet calibrated	local	~12GB	32k	llama.cpp ›
—	Eurus-2-7B-PRIME · Q8_0 PRIME-RL	not yet calibrated	local	~10GB	4k	llama.cpp ›
—	Eurus-2-7B-SFT · Q8_0 PRIME-RL	not yet calibrated	local	~10GB		llama.cpp ›
—	Flan-T5 Base · fp32 Google	not yet calibrated	local	~2.5GB		how to run ›
—	Flan-T5 Large · fp32 Google	not yet calibrated	local	~4.8GB		how to run ›
—	Flan-T5 Small · fp32 Google	not yet calibrated	local	~1.8GB		how to run ›
—	Flan-T5 XL · fp32 Google	not yet calibrated	local	~14GB		how to run ›
—	Gemma 3n E2B IT · q8_0 Google	not yet calibrated	local	~8.9GB	32k	Ollama ›
—	Gemma 3n E4B · q8_0 Google	not yet calibrated	local	~12GB	32k	Ollama ›
—	GPT-2 XL (1.5B) · fp32 OpenAI	not yet calibrated	local	~8.3GB	1k	how to run ›
—	GPT-Neo 1.3B · fp32 EleutherAI	not yet calibrated	local	~7.1GB	2k	how to run ›
—	GPT-Neo 2.7B · fp32 EleutherAI	not yet calibrated	local	~13GB	2k	how to run ›
—	Granite 4.1 30B · q4_k_m IBM	not yet calibrated	local (slow)	~19GB		Ollama ›
—	Granite 4.1 3B · q4_k_m IBM	not yet calibrated	local	~3.7GB		Ollama ›
—	Granite 4.1 8B · q8_0 IBM	not yet calibrated	local	~11GB	131k	llama.cpp ›
—	K2 Horizon 7B · gguf-q8_0 Institute of Foundation Models (MBZUAI)	not scored yet	local	~13GB	524k	llama.cpp ›
—	Llama 3.3 Nemotron Super 49B v1 · q4_k_m NVIDIA	not yet calibrated	local (slow)	~33GB		llama.cpp ›
—	Ministral 3 14B Reasoning 2512 · q4_k_m Mistral	not yet calibrated	local	~10GB	262k	llama.cpp ›
—	Ministral 3 3B Reasoning 2512 · q4_k_m Mistral	not yet calibrated	local	~3.8GB	262k	llama.cpp ›
—	Ministral 3 8B Reasoning 2512 · q4_k_m Mistral	not yet calibrated	local	~7GB	262k	llama.cpp ›
—	Muse Glimmer 30B · Q4_K_M Meta	not scored yet	local (slow)	~19GB	131k	llama.cpp ›
—	OLMo 7B · fp32 Allen Institute for AI	not yet calibrated	local (slow)	~30GB	2k	how to run ›
—	OLMo 7B Instruct · bf16 Ai2	not yet calibrated	local (slow)	~16GB		how to run ›
—	OpenChat 3.5 · q8_0 OpenChat	not yet calibrated	local	~11GB	8k	Ollama ›
—	OpenHermes 2.5 Mistral 7B · q8_0 Nous Research	not yet calibrated	local	~11GB		Ollama ›
—	Qwen2.5 0.5B Instruct · q8_0 Alibaba	not yet calibrated	local	~2.6GB	33k	Ollama ›
—	SmolLM2 360M Instruct · fp16 Hugging Face	not yet calibrated	local	~2.3GB	8k	Ollama ›

+ 245 more don't fit your machine or budget · show them

+ 1 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them

509 of 511 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