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

Best AI models for web search to run locally

A leaderboard of the best AI models for web search you can run on your own computer. Mistral Small 3.1 24B (2503) is #1 right now, followed by Llama 4 Scout (109B-A17B) and GLM-5.3-Flash. Models are ranked on web research tests like BrowseComp, with search turned on. Each model shows the download that fits and the command to run it.

68 models · scores combine 12 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 web search evaluations only. Compare scores within this task. Models without task measurements are excluded.

Filter by your hardware and budget · local only · reset

+ more

set[all sets][Arena search systems (style controlled) · 96][BrowseComp-ZH · 13][BrowseComp-ZH · 5][BrowseComp-ZH · 3][Humanity's Last Exam with tools · 3][BrowseComp with context management · 3][BrowseComp with context management · 2][Humanity's Last Exam with tools · 2][Humanity's Last Exam with tools · 2][BrowseComp · 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]

price[local][$0.05][$0.25][$1][$5][$20][any]max per M

detecting your OS…

Score vs memory to run

  60 ┤ · ● Mistral Small 3.1 24B (25…· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
     │     ∘            ∘∘                   ∘  ∘  ∘                         ·             ·       ·     ·∘
  40 ┤ · ·∘· ∘ ∘∘· · · · · · · · ·∘·∘·∘∘ · · ∘ ·∘· · · · · · · · · ·∘· · · · · · ∘ · · · · · · · · · · · · · ·
     │         ∘                  ∘       ∘          ∘                       ·             ·       ·     ·
  20 ┤ · ·∘· · · · · · ∘ · · · · ·∘· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ·
     └───────────────────────────────────────────────────────────────────────┬─────────────┬───────┬─────┬────
                                                                           500GB          1TB    1.5TB  2TB

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

#	model	task score↓	where↓	RAM↓	context↓	get it
16	Mistral Small 3.1 24B (2503) Mistral	62±3*	local	~17GB	128k	llama.cpp ›
39	Llama 4 Scout (109B-A17B) Meta	52±4*	local	~36GB	328k	llama.cpp ›
42	GLM-5.3-Flash Z.ai	52±8*†	local	~103GB	1M	llama.cpp ›
44	LongCat 2.0 (1.6T-A48B) Meituan	51±6*†	local	~2155GB	1M	how to run ›
45	GLM-5.2 Z.ai	50±3	local	~121GB	1M	llama.cpp ›
49	MiniMax-M3 MiniMax	49±3*	local	~137GB	1M	llama.cpp ›
50	Command A (03-2025) Cohere	49±5*	local	~37GB	256k	llama.cpp ›
52	Muse Glimmer 30B Meta	48±6*†	local	~19GB	131k	llama.cpp ›
55	GLM-5.1 Z.ai	46±2	local	~119GB	200k	llama.cpp ›
56	Qwen3.6 27B Alibaba	46±6*	local	~18GB	262k	LM Studio ›
60	DeepSeek V4 Pro (Preview, Apr 2026) DeepSeek	44±2	local	~604GB		llama.cpp ›
61	GLM-5 Z.ai	44±3†	local	~119GB	205k	llama.cpp ›
62	Llama 4 Maverick (402B-A17B) Meta	43±6*	local	~65GB	1M	llama.cpp ›
64	Llama 3.3 70B Instruct Meta	42±6*	local	~24GB	128k	llama.cpp ›
65	Kimi K2.6 Moonshot AI	42±3	local	~317GB	262k	llama.cpp ›
67	Qwen3.5 122B-A10B Alibaba	41±3†	local	~75GB	262k	LM Studio ›
69	Qwen3.5 397B-A17B Alibaba	41±3	local	~65GB	262k	llama.cpp ›
71	Qwen3.5 27B Alibaba	41±3†	local	~18GB	262k	LM Studio ›
74	Qwen3.5 35B-A3B Alibaba	39±3†	local	~23GB	262k	LM Studio ›
76	Gemma 4 31B Google	38±3	local	~21GB	262k	LM Studio ›
77	MiniMax-M2.7 MiniMax	38±5*	local	~75GB	205k	llama.cpp ›
78	GLM-4.7 Z.ai	38±2†	local	~58GB	205k	llama.cpp ›
80	DeepSeek V4 Flash (Preview, Apr 2026) DeepSeek	37±4*	local	~103GB		llama.cpp ›
82	Nemotron 3 Ultra 550B-A55B NVIDIA	34±4*†	local	~89GB	1M	llama.cpp ›
83	Qwen3.6 35B-A3B Alibaba	34±4*	local	~23GB	262k	LM Studio ›
85	GLM-4.6 Z.ai	31±3†	local	~58GB	205k	llama.cpp ›
87	K-EXAONE (236B-A23B) LG AI Research	30±5*	local	~152GB	262k	llama.cpp ›
88	Gemma 4 26B-A4B Google	23±8*†	local	~18GB	262k	LM Studio ›
89	GLM-4.5 Z.ai	21±4*†	local	~58GB	131k	llama.cpp ›
91	GLM-4.5-Air Z.ai	19±6*†	local	~35GB	131k	llama.cpp ›
—	DeepSeek R1 0528 DeepSeek	not yet calibrated	local	~110GB	164k	llama.cpp ›
—	DeepSeek V3.1 DeepSeek	not yet calibrated	local	~110GB	131k	llama.cpp ›
—	DeepSeek V3.1 Terminus DeepSeek	not yet calibrated	local	~230GB	164k	llama.cpp ›
—	DeepSeek V4 Flash (0731) DeepSeek	not yet calibrated	local	~92GB	1M	llama.cpp ›
—	DeepSeek V4 Pro (0813) DeepSeek	not yet calibrated	local	~1010GB	1M	llama.cpp ›
—	DeepSeek V4.1 Flash DeepSeek	not yet calibrated	local	~537GB	1M	how to run ›
—	Gemma 4 12B Google	not yet calibrated	local	~8.5GB	262k	LM Studio ›
—	Gemma 4 E2B Google	not yet calibrated	local	~5.1GB	131k	llama.cpp ›
—	Gemma 4 E4B Google	not yet calibrated	local	~6.9GB	131k	LM Studio ›
—	GLM-4.7-Flash Z.ai	not scored yet	local	~19GB	200k	LM Studio ›
—	gpt-oss-120b OpenAI	not yet calibrated	local	~67GB	131k	llama.cpp ›
—	gpt-oss-20b OpenAI	not yet calibrated	local	~14GB	131k	llama.cpp ›
—	Granite 4.2 3B IBM	not yet calibrated	local	~3.7GB	131k	LM Studio ›
—	Hunyuan Hy3 (295B-A21B) Tencent	not yet calibrated	local	~96GB	256k	llama.cpp ›
—	Hy3 preview (295B-A21B) Tencent	not yet calibrated	local	~629GB		how to run ›
—	Hy4 preview (770B-A49B) Tencent	not yet calibrated	local	~248GB	1M	llama.cpp ›
—	Inkling (975B-A41B) Thinking Machines	not yet calibrated	local	~156GB	1M	llama.cpp ›
—	Inkling Small (266B-A12B) Thinking Machines	not yet calibrated	local	~86GB	1M	llama.cpp ›
—	K2 Horizon 7B Institute of Foundation Models (MBZUAI)	not scored yet	local	~7.7GB	524k	llama.cpp ›
—	Kimi K2 Thinking Moonshot AI	not yet calibrated	local	~160GB	262k	llama.cpp ›
—	Kimi K2.5 Moonshot AI	not yet calibrated	local	~160GB	262k	llama.cpp ›
—	LFM2.5 2.6B Liquid AI	not yet calibrated	local	~3.2GB	131k	LM Studio ›
—	LFM2.5 8B-A1B Liquid AI	not yet calibrated	local	~6.5GB	128k	LM Studio ›
—	MiMo-V2-Flash (309B-A15B) Xiaomi	not yet calibrated	local	~108GB		llama.cpp ›
—	MiMo-V2.5-Pro (1T-A42B) Xiaomi	not yet calibrated	local	~163GB	1M	llama.cpp ›
—	MiniCPM5 2B OpenBMB	not yet calibrated	local	~3GB	131k	LM Studio ›
—	MiniMax-M2.1 MiniMax	not yet calibrated	local	~79GB		llama.cpp ›
—	Nemotron 3 Nano 30B-A3B NVIDIA	not scored yet	local	~20GB	262k	LM Studio ›
—	Nemotron 3 Nano 4B NVIDIA	not yet calibrated	local	~3.8GB	262k	LM Studio ›
—	Nemotron 3 Super 120B-A12B NVIDIA	not scored yet	local	~73GB	262k	LM Studio ›
—	Ornith 1.0 35B (35B-A3B) DeepReinforce	not yet calibrated	local	~23GB	262k	LM Studio ›
—	Ornith 1.5 35B-A3B DeepReinforce	not yet calibrated	local	~22GB	262k	LM Studio ›
—	Qwen3.5 2B Alibaba	not yet calibrated	local	~2.9GB	262k	llama.cpp ›
—	Qwen3.5 4B Alibaba	not yet calibrated	local	~4.3GB	262k	llama.cpp ›
—	Qwen3.5 9B Alibaba	not yet calibrated	local	~7.4GB	262k	llama.cpp ›
—	Spark-X2.5 1.7B iFLYTEK (SparkLLM)	not yet calibrated	local	~2.7GB	1M	llama.cpp ›
—	Spark-X2.5 4B iFLYTEK (SparkLLM)	not yet calibrated	local	~3.9GB	1M	LM Studio ›
—	Step 3.5 Flash (196B-A11B) StepFun	not yet calibrated	local	~118GB	256k	LM Studio ›

+ 63 more don't fit your machine or budget · show them

131 of 135 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