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

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

A leaderboard of the best AI models for coding you can run locally on a 16GB graphics card (with 32GB of system RAM). Qwen3.8 27B is #1 right now, followed by Ornith 1.5 35B-A3B and KAT-Coder-V2.5-Dev (35B-A3B). Models are ranked on coding tests like HumanEval, SWE-bench and Aider. Each model shows the download that fits and the command to run it.

317 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 coding 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]

set[all sets][Arena WebDev (Code Arena) · 549][LiveCodeBench v5 (window unspecified) · 7][HumanEval · 6][BigCodeBench · 4][BigCodeBench · 4][HumanEval · 3][LiveCodeBench (version unspecified) · 3][LiveCodeBench 2024-08–2025-05 (version unstated) · 2][HumanEval · 2][HumanEval+ · 2][LiveCodeBench 2024-08–2025-05 (version unstated) · 2][HumanEval · 2][HumanEval · 2][Tulu 2: CodexEval pass@10 · 2][HumanEval · 2]

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][any]max per M

detecting your OS…

Score vs memory to run

     │                            MiMo-V2.6-Distill-Qwen-9B        Gemma 4 26B-A4B  ● Qwen3.8 27B
     │   ·            ●─────────●──────────────────∘───∘───●───∘──∘───────∘─────∘─●─∘ ∘∘ ∘∘ ∘∘  ∘  ∘   ∘∘ ∘
  50 ┤ · ●────────────∘· · · · ·∘· · · ∘ · · ·∘· · ∘ · ∘ · ∘ · ∘ ·∘· ∘ · ·∘·∘·∘·∘· ·∘· ∘ ∘∘·∘· ·∘∘∘∘∘∘∘· · · ·
     │   ∘            ∘                ∘      ∘    ∘       ∘   ∘  ∘  ∘ ∘  ∘   ∘       ·         ∘    ∘  ∘
   0 ┤ · ∘ · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·∘· · ·
     └───┬─────────────────────────────┬───────────────────────┬──────────────────────┬───────────────────────
        2GB                           5GB                    10GB                   20GB

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

#	model	task score↓	where↓	RAM↓	context↓	get it	sources
31	Qwen3.8 27B · 4-bit Alibaba	75±1	local (slow)	~19GB	262k	Ollama ›
64	Ornith 1.5 35B-A3B · 4-bit DeepReinforce	68±3†	local (slow)	~24GB	262k	llama.cpp ›
69	KAT-Coder-V2.5-Dev (35B-A3B) · 4-bit Kuaishou (Kwaipilot)	67±3†	local (slow)	~24GB	262k	llama.cpp ›
84	Qwen3.6 27B · 4-bit Alibaba	66±2	local (slow)	~20GB	262k	Ollama ›
98	Qwen3.6 35B-A3B · 4-bit Alibaba	64±1	local (slow)	~23GB	262k	llama.cpp ›
108	Nemotron Cascade 2 30B-A3B · 4-bit NVIDIA	63±2†	local (slow)	~22GB	256k	Ollama ›
110	Qwen3.5 27B · 4-bit Alibaba	63±1	local (slow)	~19GB	262k	llama.cpp ›
113	Ornith 1.0 35B (35B-A3B) · 4-bit DeepReinforce	63±2	local (slow)	~24GB	262k	Ollama ›
116	Muse Glimmer 30B · 4-bit Meta	63±1	local (slow)	~19GB	131k	llama.cpp ›
118	Laguna XS 2.1 (33B-A3B) · 4-bit Poolside	63±3†	local (slow)	~23GB	262k	Ollama ›
119	Gemma 4 31B · 4-bit Google	62±1	local (slow)	~21GB	262k	llama.cpp ›	+1
133	Gemma 4 26B-A4B · 4-bit QAT Google	61±1	local (slow)	~18GB	262k	Ollama ›
135	MiMo-V2.6-Distill-Qwen-9B · 8-bit Xiaomi	61±3*†	local	~12GB		llama.cpp ›
136	Seed-OSS 36B Instruct · 4-bit ByteDance Seed	61±3†	local (slow)	~25GB	524k	llama.cpp ›
141	Nemotron 3.5 Lightning 30B-A3B · 4-bit NVIDIA	60±5*	local (slow)	~22GB	262k	Ollama ›
142	Gemma 4 12B · 8-bit Google	60±3	local	~15GB	262k	llama.cpp ›
143	Laguna XS.2 · 4-bit Poolside	60±1	local (slow)	~23GB	262k	llama.cpp ›
144	Devstral Small 2 24B · 4-bit Mistral	59±4*†	local (slow)	~17GB	262k	llama.cpp ›
146	Nemotron-Labs-3 Puzzle 75B-A9B · 2-bit NVIDIA	59±4*†	local (slow)	~35GB		llama.cpp ›
153	Spark-X2.5 4B · 16-bit iFLYTEK (SparkLLM)	59±2†	local	~10GB	1M	Ollama ›
159	Qwen3-Next 80B-A3B Thinking · 2-bit Alibaba	58±4*†	local (slow)	~27GB	131k	llama.cpp ›
160	Qwen3.5 35B-A3B · 4-bit Alibaba	58±1	local (slow)	~24GB	262k	llama.cpp ›
163	Phi-4-reasoning 14B · 4-bit Microsoft	58±3*†	local	~11GB	33k	Ollama ›
167	Granite 4.2 30B · 4-bit IBM	58±2†	local (slow)	~20GB	131k	llama.cpp ›
168	Ministral 3 8B Reasoning 2512 · 4-bit Mistral	58±6*†	local	~7GB	262k	llama.cpp ›
176	MiniCPM5 2B · 8-bit OpenBMB	57±2†	local	~4.3GB	131k	llama.cpp ›
178	QwQ-32B-Preview · 4-bit Alibaba	57±4*	local (slow)	~22GB	33k	llama.cpp ›
185	GLM-4.5-Air · 2-bit Z.ai	56±3	local (slow)	~35GB	131k	llama.cpp ›
189	Nemotron-Cascade-14B-Thinking · 4-bit NVIDIA	56±1†	local	~11GB		llama.cpp ›
190	Mellum2.1 12B-A2.5B Thinking · 8-bit JetBrains	56±3†	local	~15GB		llama.cpp ›
191	gpt-oss-20b · MXFP4 OpenAI	56±2	local	~15GB	131k	Ollama ›
192	Qwen3-Next 80B-A3B Instruct · 2-bit Alibaba	56±2†	local (slow)	~27GB	131k	llama.cpp ›
197	Phi-4-reasoning-plus 14B · 4-bit Microsoft	55±2	local	~13GB		Ollama ›
198	Qwen3-Coder-Next (80B-A3B) · 2-bit Alibaba	55±3*	local (slow)	~27GB	262k	llama.cpp ›
201	Qwen3-VL 32B Thinking · 4-bit Alibaba	55±3*†	local (slow)	~22GB	256k	llama.cpp ›
203	Apriel 1.6 15B Thinker · 4-bit ServiceNow	55±2*†	local	~11GB		llama.cpp ›
204	OpenReasoning-Nemotron-32B · 4-bit NVIDIA	54±2†	local (slow)	~22GB		llama.cpp ›
205	OpenThinker 32B · 4-bit Open Thoughts	54±9*†	local (slow)	~22GB	33k	llama.cpp ›
207	Qwen3-VL 30B-A3B Thinking · 4-bit Alibaba	54±3*†	local (slow)	~21GB	256k	llama.cpp ›
208	Qwen3.5 9B · 8-bit Alibaba	54±1	local	~12GB	262k	llama.cpp ›	+3
209	Qwen3 30B-A3B Thinking 2507 · 4-bit Alibaba	54±3	local (slow)	~21GB	262k	llama.cpp ›
210	Command A Reasoning (08-2025) · 2-bit Cohere	54±7*†	local (slow)	~34GB		llama.cpp ›
212	Qwen2.5 Coder 32B Instruct · 4-bit Alibaba	53±1	local (slow)	~22GB	131k	llama.cpp ›	+2
218	Qwen3 32B · 4-bit Alibaba	53±1	local (slow)	~22GB	131k	llama.cpp ›
219	Granite 4.2 8B · 8-bit IBM	53±2†	local	~11GB	131k	llama.cpp ›
222	Nemotron-Cascade-8B · 8-bit NVIDIA	53±2†	local	~11GB		llama.cpp ›
223	Granite 4.1 8B · 8-bit IBM	53±1	local	~11GB	131k	llama.cpp ›
224	Nemotron-Cascade-8B-Thinking · 8-bit NVIDIA	53±2†	local	~11GB		llama.cpp ›
225	Ministral 3 3B Reasoning 2512 · 4-bit Mistral	52±6*†	local	~3.8GB	262k	llama.cpp ›
227	Qwen2.5 32B Instruct · 4-bit Alibaba	52±1	local (slow)	~22GB	131k	llama.cpp ›
228	Qwen3 30B-A3B Instruct 2507 · 4-bit Alibaba	52±1†	local (slow)	~21GB	262k	llama.cpp ›
234	Llama 3.3 Nemotron Super 49B v1.5 · 4-bit NVIDIA	52±2†	local (slow)	~33GB	131k	llama.cpp ›
236	Olmo 3.1 32B Think · 4-bit Ai2	52±3†	local (slow)	~22GB	66k	llama.cpp ›
237	QwQ-32B · 4-bit Alibaba	52±1	local (slow)	~22GB	131k	llama.cpp ›
239	Falcon-H1R 7B · 8-bit TII	52±4*†	local	~12GB	262k	llama.cpp ›
241	Olmo 3 32B Think · 4-bit Ai2	51±3†	local (slow)	~22GB	66k	llama.cpp ›
242	Granite 4.1 30B · 4-bit IBM	51±2†	local (slow)	~19GB		Ollama ›
243	s1.1 32B · 4-bit Simple Scaling	51±9*	local (slow)	~22GB	33k	llama.cpp ›
244	Qwen3 Coder 30B-A3B Instruct · 4-bit Alibaba	51±2	local (slow)	~21GB	160k	llama.cpp ›
247	Qwen3-VL 8B Thinking · 8-bit Alibaba	51±2*†	local	~12GB	256k	Ollama ›
250	Qwen2.5 Coder 14B Instruct · 4-bit Alibaba	51±1	local	~11GB	131k	Ollama ›
255	Qwen2.5 72B Instruct · 2-bit Alibaba	51±1	local (slow)	~30GB	33k	llama.cpp ›	+2
257	Nemotron Nano 9B v2 · 8-bit NVIDIA	50±1	local	~13GB	131k	llama.cpp ›
261	Command A (03-2025) · 2-bit Cohere	50±2†	local (slow)	~37GB	256k	llama.cpp ›
265	Qwen3 4B Thinking 2507 · 8-bit Alibaba	50±6*†	local	~7.1GB	262k	Ollama ›
267	Ministral 3 14B Reasoning 2512 · 4-bit Mistral	50±4	local	~10GB	262k	llama.cpp ›
268	Phi-4 14B · 4-bit Microsoft	50±1	local	~11GB	16k	Ollama ›
269	Granite 4.2 3B · 8-bit IBM	50±2	local	~5.6GB	131k	llama.cpp ›
270	Mellum2 12B-A2.5B Thinking · 8-bit JetBrains	50±2†	local	~15GB	131k	llama.cpp ›
272	EXAONE 4.0 32B · 4-bit LG AI Research	49±3*†	local (slow)	~22GB	131k	llama.cpp ›
273	Olmo 3 7B Think · 8-bit Ai2	49±2	local	~11GB	66k	Ollama ›
274	Gemma 2 27B IT · 4-bit Google	49±2	local (slow)	~19GB	8k	llama.cpp ›
276	InternLM2.5 20B Chat · 4-bit Shanghai AI Laboratory (InternLM)	49±2	local	~15GB	33k	Ollama ›
277	Qwen3.5 4B · 8-bit Alibaba	49±1	local	~6.2GB	262k	llama.cpp ›	+1
278	Llama 3.3 70B Instruct · 2-bit Meta	49±1	local (slow)	~24GB	128k	llama.cpp ›
279	Mistral Small 3 24B (2501) · 4-bit Mistral	49±4*	local (slow)	~17GB	32k	llama.cpp ›
281	DeepSeek R1 Distill Qwen 14B · 4-bit DeepSeek	49±2	local	~11GB	131k	Ollama ›
282	Qwen2.5 14B Instruct · 4-bit Alibaba	49±1	local	~11GB	131k	Ollama ›
283	Mistral Small Instruct 22B (2409) · 4-bit Mistral AI	48±2	local	~16GB	32k	llama.cpp ›
285	Qwen3 30B-A3B (Apr 2025, hybrid) · 4-bit Alibaba	48±1†	local (slow)	~21GB	33k	llama.cpp ›
286	CodeQwen1.5 7B Chat · 8-bit Alibaba	48±1	local	~11GB	64k	Ollama ›
287	OpenCoder-8B-Instruct · 8-bit INF (infly)	48±1	local	~10GB		llama.cpp ›
288	DeepSeek R1 Distill Llama 70B · 2-bit DeepSeek	48±2	local (slow)	~30GB	131k	llama.cpp ›
289	Llama 3.1 70B Instruct · 2-bit Meta	48±1	local (slow)	~27GB	128k	llama.cpp ›	+1
290	Qwen3 8B · 8-bit Alibaba	48±2	local	~12GB	33k	Ollama ›
292	Qwen2.5 Coder 7B Instruct · 8-bit Alibaba	48±1	local	~12GB	131k	Ollama ›
293	DeepSeek R1 Distill Qwen 32B · 4-bit DeepSeek	48±2	local (slow)	~22GB	131k	llama.cpp ›
295	DeepSeek R1 0528 Qwen3 8B · 8-bit DeepSeek	47±4	local	~12GB	128k	Ollama ›
296	Qwen3-VL 4B Thinking · 8-bit Alibaba	47±2*†	local	~6.9GB	256k	Ollama ›
297	Nemotron 3 Nano 30B-A3B · 4-bit NVIDIA	47±3†	local (slow)	~22GB	262k	Ollama ›
299	DeepSeek LLM 67B Chat · 2-bit DeepSeek	47±2	local (slow)	~31GB	4k	llama.cpp ›
300	OpenThinker3 7B · 8-bit Open Thoughts	47±2	local	~10GB	33k	llama.cpp ›
302	Qwen2 72B Instruct · 2-bit Alibaba	47±1	local (slow)	~33GB	131k	llama.cpp ›
304	Phi-3 Medium 128K Instruct · 4-bit Microsoft	47±2	local	~11GB	128k	Ollama ›
305	OpenCodeInterpreter-DS-33B · 4-bit M-A-P	47±1	local (slow)	~22GB		llama.cpp ›
307	Gemma 4 E4B · 8-bit Google	47±1	local	~9.9GB	131k	llama.cpp ›
308	DeepSeek-Coder 33B Instruct · 4-bit DeepSeek	46±1	local (slow)	~22GB	16k	llama.cpp ›
309	Codestral 22B v0.1 · 4-bit Mistral AI	46±1	local	~16GB	32k	llama.cpp ›
310	WizardCoder 33B V1.1 · 4-bit WizardLM	46±1	local (slow)	~22GB	16k	llama.cpp ›
311	Gemma 2 9B IT · 8-bit Google	46±2	local	~14GB	8k	Ollama ›
313	Qwen2.5 7B Instruct · 8-bit Alibaba	46±1	local	~12GB	131k	Ollama ›
314	Mistral NeMo Instruct (2407) · 8-bit Mistral	46±2	local	~15GB	128k	llama.cpp ›
316	Llama 3 70B Instruct · 2-bit Meta	46±1	local (slow)	~24GB	8k	llama.cpp ›
317	OpenCodeInterpreter-DS-6.7B · 8-bit M-A-P	46±1	local	~9.1GB		llama.cpp ›
320	speechless-codellama-34b-v2.0 · 4-bit uukuguy	46±1	local (slow)	~23GB		llama.cpp ›
321	Magicoder-S-DS 6.7B · 8-bit UIUC (ise-uiuc)	46±1	local	~9.1GB		llama.cpp ›
322	Artigenz-Coder-DS-6.7B · 4-bit Artigenz	46±1	local	~5.8GB		llama.cpp ›
325	Yi 1.5 9B Chat · 8-bit 01.AI	45±2	local	~13GB	4k	Ollama ›
326	Nemotron 3 Nano 4B · 8-bit NVIDIA	45±2	local	~7GB	262k	Ollama ›
327	Pixtral 12B (2409) · 8-bit Mistral	45±3*	local	~15GB		llama.cpp ›
329	Granite 4.1 3B · 4-bit IBM	45±2†	local	~3.7GB		Ollama ›
330	XwinCoder-34B · 4-bit Xwin-LM	45±1	local (slow)	~23GB		llama.cpp ›
331	StarChat2 15B v0.1 · 4-bit Hugging Face	45±1	local	~12GB	16k	llama.cpp ›
332	Code Millenials 34B · 4-bit Bud Ecosystem	45±1	local (slow)	~23GB		llama.cpp ›
333	WizardLM 70B V1.0 · 2-bit Microsoft	45±2	local (slow)	~32GB		llama.cpp ›
334	DeepSeek-Coder 6.7B Instruct · 8-bit DeepSeek	45±1	local	~9.1GB	16k	Ollama ›
335	AceReason-Nemotron-1.0-14B · 4-bit NVIDIA	45±2†	local	~11GB		llama.cpp ›
336	WaveCoder-Ultra-6.7B · 8-bit Microsoft	45±1	local	~9.1GB		llama.cpp ›
338	Yi-Coder 9B Chat · 8-bit 01.AI	45±1	local	~13GB	128k	Ollama ›
339	MiniCPM3-4B · 4-bit OpenBMB	45±2*	local	~4.1GB	32k	llama.cpp ›
340	Gemma 3 27B · 4-bit Google	44±1	local (slow)	~19GB	131k	llama.cpp ›
341	Ministral 3 14B · 4-bit Mistral	44±4*	local	~11GB	262k	Ollama ›
342	OpenThinker2 7B · 4-bit Open Thoughts	44±9*†	local	~6.4GB	33k	llama.cpp ›
343	DeepSeek-Coder 7B Instruct v1.5 · 8-bit DeepSeek	44±1	local	~9.2GB		llama.cpp ›
345	Qwen3 4B · 8-bit Alibaba	44±2	local	~7.1GB	33k	Ollama ›
346	Gemma 3 12B · 8-bit Google	44±2	local	~15GB	131k	llama.cpp ›
347	Yi 1.5 34B Chat · 4-bit 01.AI	44±2	local (slow)	~23GB	4k	llama.cpp ›
349	Qwen3-VL 32B Instruct · 4-bit Alibaba	44±2*†	local (slow)	~22GB	256k	llama.cpp ›
350	OpenChat 3.5 0106 · 8-bit OpenChat	44±1	local	~11GB	8k	Ollama ›
351	Athene-70B · 2-bit Nexusflow	43±3*	local (slow)	~29GB		llama.cpp ›
352	Granite 34B Code Instruct (8K) · 4-bit IBM	43±1	local (slow)	~22GB		Ollama ›
353	Phind CodeLlama 34B v2 · 4-bit Phind	43±1	local (slow)	~23GB	16k	Ollama ›
354	WhiteRabbitNeo-33B-v1 · 4-bit WhiteRabbitNeo	43±1	local (slow)	~22GB		llama.cpp ›
355	Magicoder-S-CL-7B · 8-bit UIUC (ise-uiuc)	43±1	local	~9.1GB		llama.cpp ›
356	Qwen2.5 Coder 3B Instruct · 4-bit Alibaba	43±1	local	~3.7GB	33k	llama.cpp ›
357	Code Llama 70B Instruct · 2-bit Meta	43±1	local (slow)	~28GB	16k	llama.cpp ›
358	Qwen3-VL 30B-A3B Instruct · 4-bit Alibaba	43±2*†	local (slow)	~21GB	256k	llama.cpp ›
359	speechless-coder-ds-6.7b · 8-bit uukuguy	43±1	local	~9.1GB		llama.cpp ›
360	Spark-X2.5 1.7B · 16-bit iFLYTEK (SparkLLM)	43±3†	local	~5.1GB	1M	Ollama ›
362	WizardCoder Python 34B V1.0 · 4-bit WizardLM	43±1	local (slow)	~23GB	16k	Ollama ›
363	Phi-4-mini 3.8B · 8-bit Microsoft	43±2†	local	~6.9GB	128k	Ollama ›
365	AceReason-Nemotron-1.1-7B · 8-bit NVIDIA	42±2†	local	~10GB		llama.cpp ›
367	DeepSeek-Coder-V2 Lite Instruct (16B-A2.4B) · 4-bit DeepSeek	42±2	local	~12GB	128k	llama.cpp ›
369	Qwen3 4B Instruct 2507 · 8-bit Alibaba	42±2	local	~7.1GB	262k	Ollama ›
373	Mellum2 12B-A2.5B Instruct · 8-bit JetBrains	42±4*†	local	~15GB	131k	llama.cpp ›
374	Llama 3.1 8B Instruct · 8-bit Meta	42±1	local	~12GB	128k	Ollama ›	+2
375	Llama 3.2 11B Vision Instruct · 4-bit Meta	42±3*	local	~8.8GB	128k	Ollama ›
376	speechless-starcoder2-15b · 4-bit uukuguy	42±1	local	~12GB		llama.cpp ›
377	Granite 3.0 8B Instruct · 8-bit IBM	42±2	local	~12GB	4k	Ollama ›
378	Qwen2-VL 7B Instruct · 8-bit Alibaba	42±3*	local	~11GB		llama.cpp ›
379	Phi-3-mini-128k-instruct (Apr 2024) · 16-bit Microsoft	42±2	local	~9.5GB		how to run ›
380	Qwen2.5 3B Instruct · 8-bit Alibaba	42±2	local	~5.9GB	33k	Ollama ›
381	Qwen1.5 72B Chat · 2-bit Alibaba	41±1	local (slow)	~31GB	32k	llama.cpp ›
382	Phi-3-mini-4k-instruct (June 2024 update) · 8-bit Microsoft	41±3*	local	~5.8GB		llama.cpp ›
383	Qwen3-VL 8B Instruct · 8-bit Alibaba	41±2*†	local	~12GB	256k	Ollama ›
384	StarCoder2-15B-Instruct-v0.1 · 4-bit BigCode	41±1	local	~12GB	16k	llama.cpp ›
386	Granite 8B Code Instruct (4K) · 4-bit IBM	41±2†	local	~6.6GB		llama.cpp ›
387	Granite 20B Code Instruct (8K) · 4-bit IBM	41±2†	local	~15GB		llama.cpp ›
388	Phi-3.5-mini-instruct · 8-bit Microsoft	41±2	local	~6.8GB	128k	Ollama ›
389	Falcon3 7B Instruct · 8-bit TII	41±2	local	~11GB	32k	Ollama ›
390	Qwen3-VL 4B Instruct · 8-bit Alibaba	41±2*†	local	~6.9GB	256k	Ollama ›
392	Code Llama 70B Python · 2-bit Meta	40±1	local (slow)	~28GB	16k	llama.cpp ›
393	Dolphin 2.6 Mixtral 8x7B · 4-bit Cognitive Computations	40±1	local (slow)	~29GB		llama.cpp ›
394	Command R (08-2024) · 4-bit Cohere	40±1	local (slow)	~22GB	128k	llama.cpp ›
395	DeepSeek-Coder 1.3B Instruct · 8-bit DeepSeek	40±1	local	~3GB	16k	Ollama ›
397	GLM-4.7-Flash · 4-bit Z.ai	40±6*	local (slow)	~21GB	200k	llama.cpp ›
398	Olmo 3 7B Instruct · 8-bit Ai2	39±2	local	~11GB	66k	Ollama ›
399	OpenChat 3.5 · 8-bit OpenChat	39±3*†	local	~11GB	8k	Ollama ›
400	Llama 3 8B Instruct · 8-bit Meta	39±1	local	~12GB	8k	Ollama ›
401	Code Llama 70B (base) · 2-bit Meta	39±1	local (slow)	~28GB	16k	llama.cpp ›
402	Qwen2 7B Instruct · 8-bit Alibaba	39±3*	local	~12GB	131k	Ollama ›
403	Falcon3 10B Instruct · 8-bit TII	39±2	local	~15GB	32k	Ollama ›
404	Code Tulu 2 13B · 16-bit Allen Institute for AI	39±2*†	local (slow)	~29GB		how to run ›
405	WizardCoder Python 13B V1.0 · 4-bit WizardLM	39±3	local	~9.8GB	16k	Ollama ›
406	Qwen2.5 Coder 1.5B Instruct · 8-bit Alibaba	39±1	local	~3.5GB	33k	llama.cpp ›
407	Gemma 4 E2B · 8-bit Google	39±2	local	~6.8GB	131k	llama.cpp ›
408	CodeGemma 7B IT (v1.0) · 8-bit Google	39±1	local	~11GB		Ollama ›
409	LFM2.5 2.6B · 8-bit Liquid AI	38±2	local	~4.5GB	131k	llama.cpp ›
410	Granite 34B Code Base (8K) · 4-bit IBM	38±1†	local (slow)	~24GB		llama.cpp ›
412	speechless-starcoder2-7b · 8-bit uukuguy	38±1	local	~9.5GB		llama.cpp ›
414	Code Llama 34B Python · 4-bit Meta	38±1	local (slow)	~23GB	16k	Ollama ›
415	WizardCoder-15B-V1.0 · 4-bit WizardLM	38±1	local	~12GB		llama.cpp ›
416	Qwen2.5 1.5B Instruct · 8-bit Alibaba	37±2	local	~4GB	33k	Ollama ›
417	Hermes 3 Llama 3.1 8B · 8-bit Nous Research	37±2*	local	~12GB	128k	Ollama ›
418	speechless-coding-7b-16k-tora · 16-bit uukuguy	37±1	local (slow)	~16GB		how to run ›
419	Qwen3 1.7B · 8-bit Alibaba	37±2*	local	~4.6GB	33k	Ollama ›
420	OpenHermes 2.5 Mistral 7B · 8-bit Nous Research	37±3*	local	~11GB		Ollama ›
421	Gemma 3 4B · 8-bit Google	37±2	local	~7.4GB	131k	Ollama ›
422	Code Llama 34B Instruct · 4-bit Meta	37±1	local (slow)	~23GB	16k	llama.cpp ›
423	Command R7B (12-2024) · 8-bit Cohere	37±2†	local	~12GB	128k	Ollama ›
424	OpenHermes-2.5-Code-290k-13B · 4-bit ajibawa-2023	37±1	local	~9.8GB		llama.cpp ›
425	WizardLM 13B V1.2 · 4-bit Microsoft	36±3	local	~9.8GB		llama.cpp ›
426	Qwen1.5 14B Chat · 4-bit Alibaba	36±1*	local	~11GB	32k	llama.cpp ›
427	Granite 3.0 2B Instruct · 8-bit IBM	36±2	local	~5.3GB	4k	Ollama ›
428	DeepSeek-Coder 33B Base · 4-bit DeepSeek	36±1	local (slow)	~22GB	16k	llama.cpp ›
429	Qwen3-VL 2B Thinking · 8-bit Alibaba	36±3*†	local	~4.2GB	256k	Ollama ›
430	CodeQwen1.5 7B (base) · 8-bit Alibaba	36±1	local	~9.6GB	66k	llama.cpp ›
431	Code Llama 34B (base) · 4-bit Meta	36±1	local (slow)	~23GB	16k	Ollama ›
432	Llama 3.2 3B Instruct · 8-bit Meta	36±1	local	~6.1GB	131k	Ollama ›
433	Granite 3B Code Instruct (2K) · 4-bit IBM	36±2†	local	~3.7GB		llama.cpp ›
434	Phi-2 (2.7B) · 4-bit Microsoft	35±1	local	~7.3GB	2k	how to run ›
435	Falcon3 3B Instruct · 8-bit TII	35±3	local	~6.1GB	32k	Ollama ›
436	SmolLM3 3B · 8-bit Hugging Face	35±6*†	local	~5.9GB	66k	llama.cpp ›
437	Tulu 2 DPO 70B · 2-bit Ai2	35±2*†	local (slow)	~32GB		llama.cpp ›
438	Qwen-14B-Chat · 4-bit Alibaba	35±1*	local	~11GB		llama.cpp ›
439	Code Llama 13B Instruct · 4-bit Meta	35±1	local	~9.8GB	16k	Ollama ›
440	Code Tulu 2 7B · 16-bit Allen Institute for AI	35±2*†	local (slow)	~16GB		how to run ›
441	LFM2.5 8B-A1B · 8-bit Liquid AI	35±2	local	~13GB	128k	Ollama ›
442	DeepSeek-V2 Lite Chat (16B-A2.4B) · 4-bit DeepSeek	35±2	local	~12GB	32k	llama.cpp ›
443	Tulu 2 70B · 2-bit Allen Institute for AI	35±2*†	local (slow)	~28GB		llama.cpp ›
444	Granite 8B Code Base (4K) · 4-bit IBM	35±1†	local	~6.6GB		llama.cpp ›
445	DeepSeek-Coder 6.7B Base · 8-bit DeepSeek	35±1	local	~9.1GB	16k	Ollama ›
446	Mixtral 8x7B Instruct v0.1 · 4-bit Mistral	34±1	local (slow)	~29GB	32k	llama.cpp ›
447	MistralHermes-CodePro-7B-v1 · 8-bit beowolx	34±1	local	~9.6GB		llama.cpp ›
448	speechless-code-mistral-7b-v1.0 · 4-bit uukuguy	34±1	local (slow)	~17GB		how to run ›
449	Code Llama 13B Python · 4-bit Meta	34±1	local	~9.8GB	16k	Ollama ›
450	StarCoder2 15B · 4-bit BigCode	34±1	local	~12GB	16k	Ollama ›
451	Mixtral 8x7B v0.1 (base) · 4-bit Mistral AI	34±1	local (slow)	~29GB	32k	llama.cpp ›
452	Xwin-LM 70B v0.1 · 2-bit Xwin-LM	34±2*	local (slow)	~28GB		llama.cpp ›
453	CodeGemma 7B (v1.0) · 8-bit Google	34±1	local	~11GB		Ollama ›
454	Qwen1.5 7B Chat · 8-bit Alibaba	34±1*	local	~12GB	32k	Ollama ›
455	EXAONE 3.5 2.4B Instruct · 8-bit LG AI Research	34±3	local	~5.1GB	33k	Ollama ›
456	Granite 20B Code Base (8K) · 4-bit IBM	33±2†	local	~15GB		llama.cpp ›
457	Code Llama 7B Instruct · 8-bit Meta	33±1	local	~9.1GB	16k	Ollama ›
458	Code Llama 7B Python · 8-bit Meta	33±1	local	~9.1GB	16k	Ollama ›
459	Code Llama 13B (base) · 4-bit Meta	33±1	local	~9.8GB	16k	Ollama ›
460	SOLAR 10.7B Instruct v1.0 · 8-bit Upstage	33±1	local	~14GB	4k	llama.cpp ›
461	Qwen2.5 Coder 0.5B Instruct · 4-bit Alibaba	33±1	local	~2GB	33k	llama.cpp ›
462	Gemma 1.1 7B IT · 8-bit Google	32±1	local	~13GB	8k	Ollama ›
463	Gemma 2 2B IT · 8-bit Google	32±2	local	~5.3GB	8k	Ollama ›
464	DeepSeek LLM 7B Chat · 4-bit DeepSeek	32±3*†	local	~5.7GB		Ollama ›
465	SmolLM2 1.7B Instruct · 8-bit Hugging Face	32±4*†	local	~4.2GB	8k	Ollama ›
466	Mistral 7B Instruct v0.2 · 8-bit Mistral	32±1	local	~11GB	32k	Ollama ›
467	Llama 3.2 1B Instruct · 4-bit Meta	31±2	local	~5.2GB	131k	Ollama ›
468	xDAN-L1-Chat-RL-v1 · 8-bit xDAN-AI	31±1	local	~9.6GB		llama.cpp ›
469	Granite 3.0 3B-A800M Instruct · 8-bit IBM	31±2*†	local	~5.3GB	4k	llama.cpp ›
470	Code Llama 7B (base) · 8-bit Meta	31±1	local	~9.1GB	16k	Ollama ›	+1
471	Llama 2 70B (base) · 2-bit Meta	30±1	local (slow)	~32GB	4k	llama.cpp ›
472	Granite 3B Code Base (2K) · 4-bit IBM	30±2†	local	~3.7GB		llama.cpp ›
473	StarCoder2 7B · 8-bit BigCode	30±1	local	~9.5GB	16k	Ollama ›
474	Qwen3-VL 2B Instruct · 8-bit Alibaba	30±3*†	local	~4.2GB	256k	Ollama ›
475	StarCoder 15.5B · 4-bit BigCode	30±1	local	~12GB		llama.cpp ›
476	Gemma 7B (base) · 8-bit Google	30±1	local	~11GB	8k	Ollama ›
477	Llama 3 8B (base) · 8-bit Meta	30±1	local	~10GB		llama.cpp ›
478	StarCoder2 3B · 8-bit BigCode	29±1	local	~4.9GB	16k	Ollama ›
479	StarCoderBase 15.5B · 4-bit BigCode	29±1	local	~12GB		llama.cpp ›
480	Mistral 7B Instruct v0.1 · 8-bit Mistral	29±1*	local	~9.6GB	8k	llama.cpp ›
481	Llama 2 70B Chat · 2-bit Meta	28±2*	local (slow)	~32GB	4k	llama.cpp ›
483	DeepSeek-Coder 1.3B Base · 8-bit DeepSeek	28±1	local	~3GB	16k	Ollama ›
484	Stable Code 3B · 8-bit Stability AI	28±1	local	~4.7GB	16k	Ollama ›
485	OpenThinker 7B · 8-bit Open Thoughts	28±8*†	local	~10GB	33k	llama.cpp ›
486	Mistral 7B v0.1 (base) · 8-bit Mistral AI	28±1	local	~9.6GB	8k	llama.cpp ›
487	Gemma 7B IT · 8-bit Google	28±1	local	~13GB	8k	Ollama ›
488	Zephyr 7B beta · 8-bit Hugging Face	28±1	local	~11GB		Ollama ›
489	DeepSeek R1 Distill Qwen 7B · 8-bit DeepSeek	27±3	local	~12GB	131k	Ollama ›
490	DeepSeek R1 Distill Llama 8B · 8-bit DeepSeek	27±4	local	~12GB	131k	Ollama ›
491	Tulu 2 13B · 4-bit Allen Institute for AI	27±2*†	local	~9.8GB		llama.cpp ›
492	Tulu 2 DPO 13B · 4-bit Allen Institute for AI	27±2*†	local	~9.8GB		llama.cpp ›
493	Granite 3.0 1B-A400M Instruct · 8-bit IBM	27±2*†	local	~3GB	4k	llama.cpp ›
494	CodeGemma 2B (v1.0) · 8-bit Google	27±1	local	~4.3GB		Ollama ›
496	Falcon 40B (base) · 4-bit TII	26±2*	local (slow)	~27GB	2k	Ollama ›
497	StarCoderBase 7B · 4-bit BigCode	26±1	local	~6GB		Ollama ›
498	Falcon3 1B Instruct · 8-bit TII	26±4	local	~4.1GB	8k	Ollama ›
499	Gemma 3 1B · 4-bit Google	25±2†	local	~2.7GB	32k	Ollama ›
500	Gemma 2B (base) · 8-bit Google	25±1	local	~4.3GB	8k	Ollama ›
501	Qwen3.5 2B · 8-bit Alibaba	24±3	local	~4.9GB	262k	Ollama ›
502	Gemma 1.1 2B IT · 8-bit Google	24±1	local	~5.2GB	8k	Ollama ›
503	Tulu 2 DPO 7B · 8-bit Allen Institute for AI	23±2*†	local	~9.1GB		llama.cpp ›
504	StarCoderBase 3B · 4-bit BigCode	23±1	local	~14GB		how to run ›
506	Llama 2 13B (base) · 4-bit Meta	22±1	local	~9.8GB	4k	llama.cpp ›
507	Tulu 2 7B · 8-bit Allen Institute for AI	22±2*†	local	~9.1GB		llama.cpp ›
508	Gemma 2B IT · 8-bit Google	22±1	local	~5.2GB	8k	Ollama ›
510	Llama 2 13B Chat · 4-bit Meta	21±2*	local	~10GB	4k	Ollama ›
512	OpenAssistant SFT-4 Pythia 12B · 4-bit LAION (OpenAssistant)	20±2*	local	~9.5GB		llama.cpp ›
513	Phi-1 (1.3B) · 4-bit Microsoft	19±5*	local	~4.5GB	2k	how to run ›
514	GPT-J 6B · 4-bit EleutherAI	19±1*	local	~14GB	2k	how to run ›
515	Llama 2 7B (base) · 8-bit Meta	19±1	local	~9.1GB	4k	llama.cpp ›
517	Qwen3 0.6B · 8-bit Alibaba	17±3*	local	~3GB	33k	Ollama ›
518	Llama 2 7B Chat · 8-bit Meta	16±2*	local	~11GB	4k	Ollama ›
519	Molmo 7B-D (0924) · 4-bit Ai2	14±5*	local (slow)	~35GB		how to run ›
520	GPT-Neo 2.7B · 4-bit EleutherAI	14±1*	local	~13GB	2k	how to run ›
521	DeepSeek R1 Distill Qwen 1.5B · 8-bit DeepSeek	14±5	local	~4.3GB	131k	Ollama ›
522	ChatGLM-6B · 4-bit Zhipu AI (THUDM)	12±3*	local (slow)	~16GB		how to run ›
523	StableLM-Base-Alpha 7B · 4-bit Stability AI	5±2	local (slow)	~35GB	4k	how to run ›
524	Zyte-1B · 4-bit venkycs (aihub-app)	3±2	local	~3.8GB		how to run ›
—	BLOOM 1.1B · fp16 BigScience	not yet calibrated	local	~3.7GB	2k	how to run ›
—	BLOOM 1.7B · fp16 BigScience	not yet calibrated	local	~5.1GB	2k	how to run ›
—	BLOOM 3B · fp16 BigScience	not yet calibrated	local	~7.8GB	2k	how to run ›
—	BLOOM 560M · fp16 BigScience	not yet calibrated	local	~2.7GB	2k	how to run ›
—	BLOOM 7.1B · Q8_0 BigScience	not yet calibrated	local	~11GB	2k	llama.cpp ›
—	EXAONE Deep 2.4B · q4_k_m LG AI Research	not yet calibrated	local	~3.2GB	33k	llama.cpp ›
—	EXAONE Deep 32B · q4_k_m LG AI Research	not yet calibrated	local (slow)	~22GB	33k	Ollama ›
—	EXAONE Deep 7.8B · q4_k_m LG AI Research	not yet calibrated	local	~6.5GB	33k	Ollama ›
—	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 ›
—	Granite 4.0 1B · bf16 IBM	not yet calibrated	local	~6.1GB	128k	Ollama ›
—	Granite 4.0 350M · bf16 IBM	not yet calibrated	local	~2.9GB	32k	Ollama ›
—	Granite 4.0 H 1B · q8_0 IBM	not yet calibrated	local	~3.2GB	131k	Ollama ›
—	Granite 4.0 H 350M · q8_0 IBM	not yet calibrated	local	~1.9GB	33k	Ollama ›
—	Granite 4.0 H Micro 3B · q4_k_m IBM	not yet calibrated	local	~3.5GB	131k	llama.cpp ›
—	Granite 4.0 H Small (32B-A9B) · Q4_K_M IBM	not yet calibrated	local (slow)	~22GB	131k	llama.cpp ›
—	Granite 4.0 H Tiny (7B-A1B) · q4_k_m IBM	not yet calibrated	local	~5.9GB	131k	llama.cpp ›
—	Granite 4.0 Micro 3B · Q8_0 IBM	not yet calibrated	local	~5.3GB	131k	llama.cpp ›
—	Jamba 1.5 Mini · Q4_K_M AI21 Labs	not yet calibrated	local (slow)	~34GB		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 scored yet	local (slow)	~33GB		llama.cpp ›
—	Magistral Small 1.0 (2506) · Q4_K_M Mistral	not yet calibrated	local (slow)	~17GB	128k	llama.cpp ›
—	Magistral Small 1.1 (2507) · Q4_K_M Mistral	not yet calibrated	local (slow)	~17GB	40k	llama.cpp ›
—	Magistral Small 1.2 · gguf-q4_k_m Mistral	not yet calibrated	local (slow)	~17GB	131k	llama.cpp ›
—	Ministral 8B (2410) · Q8_0 Mistral	not yet calibrated	local	~10GB	128k	llama.cpp ›
—	Mistral 7B Instruct v0.3 · q8_0 Mistral	not yet calibrated	local	~11GB	32k	Ollama ›
—	Mistral Small 3.1 24B (2503) · Q4_K_M Mistral	not scored yet	local (slow)	~17GB	128k	llama.cpp ›
—	Mistral Small 3.2 24B · Q4_K_M Mistral	not scored yet	local (slow)	~17GB	128k	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 ›
—	Qwen2 0.5B Instruct · q8_0 Alibaba	not yet calibrated	local	~2.6GB	32k	Ollama ›
—	Qwen2 1.5B Instruct · q8_0 Alibaba	not yet calibrated	local	~4GB	32k	Ollama ›

+ 255 more don't fit your machine or budget · show them

+ 9 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them

572 of 598 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