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

Best AI models for web search to run locally on a 32GB graphics card

A leaderboard of the best AI models for web search you can run locally on a 32GB graphics card (with 32GB of system RAM). Mistral Small 3.1 24B (2503) is #1 right now, followed by Llama 4 Scout (109B-A17B) and Command A (03-2025). 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.

31 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.

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

+ more

card[RTX 5090][RTX 5000 Ada][RTX PRO 4500 Blackwell]

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]

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

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 ┤ · · · · ·∘· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · ∘ · · · · ·
     └───────────────────────┬───────────────────────────┬──────────────────────┬──────────────────┬──────────
                           20GB                        25GB                   30GB               35GB

#	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) · 8-bit Mistral	62±3*	local	~28GB	128k	llama.cpp ›
39	Llama 4 Scout (109B-A17B) · 2-bit Meta	52±4*	local (slow)	~36GB	328k	llama.cpp ›
50	Command A (03-2025) · 2-bit Cohere	49±5*	local (slow)	~37GB	256k	llama.cpp ›
52	Muse Glimmer 30B · 4-bit Meta	48±6*†	local	~21GB	131k	Ollama ›
56	Qwen3.6 27B · 4-bit Alibaba	46±6*	local	~22GB	262k	llama.cpp ›
64	Llama 3.3 70B Instruct · 2-bit Meta	42±6*	local	~24GB	128k	llama.cpp ›
71	Qwen3.5 27B · 4-bit Alibaba	41±3†	local	~20GB	262k	Ollama ›
74	Qwen3.5 35B-A3B · 4-bit Alibaba	39±3†	local	~25GB	262k	Ollama ›
76	Gemma 4 31B · 4-bit Google	38±3	local	~21GB	262k	Ollama ›
83	Qwen3.6 35B-A3B · 4-bit Alibaba	34±4*	local	~24GB	262k	Ollama ›
88	Gemma 4 26B-A4B · 8-bit Google	23±8*†	local	~30GB	262k	llama.cpp ›
91	GLM-4.5-Air · 2-bit Z.ai	19±6*†	local (slow)	~35GB	131k	llama.cpp ›
—	Gemma 4 12B · Q8_0 Google	not yet calibrated	local	~15GB	262k	llama.cpp ›
—	Gemma 4 E2B · Q8_0 Google	not yet calibrated	local	~6.8GB	131k	llama.cpp ›
—	Gemma 4 E4B · Q8_0 Google	not yet calibrated	local	~9.9GB	131k	llama.cpp ›
—	GLM-4.7-Flash · q4_k_m Z.ai	not scored yet	local	~22GB	200k	Ollama ›
—	gpt-oss-20b · mxfp4 OpenAI	not yet calibrated	local	~15GB	131k	Ollama ›
—	Granite 4.2 3B · Q8_0 IBM	not yet calibrated	local	~5.6GB	131k	llama.cpp ›
—	K2 Horizon 7B · gguf-q8_0 Institute of Foundation Models (MBZUAI)	not scored yet	local	~13GB	524k	llama.cpp ›
—	LFM2.5 2.6B · q8_0 Liquid AI	not yet calibrated	local	~4.5GB	131k	llama.cpp ›
—	LFM2.5 8B-A1B · q8_0 Liquid AI	not yet calibrated	local	~13GB	128k	Ollama ›
—	MiniCPM5 2B · q8_0 OpenBMB	not yet calibrated	local	~4.3GB	131k	llama.cpp ›
—	Nemotron 3 Nano 30B-A3B · Q4_K_M NVIDIA	not scored yet	local	~25GB	262k	llama.cpp ›
—	Nemotron 3 Nano 4B · q8_0 NVIDIA	not yet calibrated	local	~7GB	262k	Ollama ›
—	Ornith 1.0 35B (35B-A3B) · Q4_K_M DeepReinforce	not yet calibrated	local	~24GB	262k	llama.cpp ›
—	Ornith 1.5 35B-A3B · q4_k_m DeepReinforce	not yet calibrated	local	~26GB	262k	Ollama ›
—	Qwen3.5 2B · q8_0 Alibaba	not yet calibrated	local	~4.9GB	262k	Ollama ›
—	Qwen3.5 4B · Q8_0 Alibaba	not yet calibrated	local	~6.2GB	262k	llama.cpp ›
—	Qwen3.5 9B · Q8_0 Alibaba	not yet calibrated	local	~12GB	262k	llama.cpp ›
—	Spark-X2.5 1.7B · bf16 iFLYTEK (SparkLLM)	not yet calibrated	local	~5.1GB	1M	Ollama ›
—	Spark-X2.5 4B · bf16 iFLYTEK (SparkLLM)	not yet calibrated	local	~10GB	1M	Ollama ›

+ 100 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