RTX 3060 TESTS

What actually runs
on 12 GB of VRAM.

Scout’s own test server downloads open models from the Ollama library and times them on one NVIDIA RTX 3060, every morning from 4:00 to 8:30 AM Arizona time. Models that don’t fit in 12 GB still run, partly from system memory, so you can see how much slower that is.

223models timed
163fit in 12 GB
60don’t fit, still ran
20couldn’t run
Oct 31, 2026last updated

The test machine

Compare these with your own computer. When a model fits in VRAM, speed mostly follows the card’s memory bandwidth; when it doesn’t, the processor and system memory set the pace.

Graphics card
NVIDIA GeForce RTX 3060
Video memory (VRAM)
12 GB GDDR6360 GB/s · 192-bit bus
Processor
Intel Core i5-7600K
System memory (RAM)
32 GB DDR42,133 MT/s
Software
Ollama 0.40.13NVIDIA driver 580.178.04 · Ubuntu 26.04 LTS
Board & storage
Gigabyte Z270-HD3256 GB SSD for models

223 of 223 tested models · not a quality ranking

ModelWriting speedReading promptFits in 12 GB?DownloadVRAM usedTested
smollm2:135m ↗134.52M parameters · F16406.5 tok/s32,831 tok/sFits0.3 GB0.5 GBOct 22, 2026
qwen:0.5b ↗620M parameters · Q4_0357.2 tok/s17,033 tok/sFits0.4 GB0.6 GBOct 22, 2026
qwen:0.5b-text ↗620M parameters · Q4_0353.8 tok/s17,092 tok/sFits0.4 GB0.6 GBOct 22, 2026
gemma3:270m-it-qat ↗268.10M parameters · Q4_0347.7 tok/s27,626 tok/sFits0.2 GB0.4 GBOct 21, 2026
gemma3:270m ↗268.10M parameters · Q8_0340.2 tok/s28,000 tok/sFits0.3 GB0.4 GBOct 21, 2026
qwen2:0.5b ↗494.03M parameters · Q4_0337.5 tok/s21,246 tok/sFits0.4 GB0.5 GBOct 22, 2026
functiongemma:latest ↗268.10M parameters · Q8_0336.8 tok/s27,783 tok/sFits0.3 GB0.4 GBOct 21, 2026
granite4:350m-h ↗340.33M parameters · Q8_0327 tok/s5,950 tok/sFits0.4 GB0.5 GBOct 22, 2026
tinyllama:latest ↗1B parameters · Q4_0315.1 tok/s11,353 tok/sFits0.6 GB0.8 GBOct 22, 2026
tinyllama:v0.6 ↗1B parameters · Q4_0313.6 tok/s11,343 tok/sFits0.6 GB0.8 GBOct 22, 2026
qwen2.5-coder:0.5b ↗494.03M parameters · Q4_K_M309.7 tok/s20,371 tok/sFits0.4 GB0.6 GBOct 21, 2026
qwen2.5:0.5b-base ↗494.03M parameters · Q4_K_M298 tok/s20,339 tok/sFits0.4 GB0.6 GBOct 21, 2026
qwen2.5-coder:0.5b-base ↗494.03M parameters · Q8_0292.4 tok/s21,242 tok/sFits0.5 GB0.7 GBOct 22, 2026
qwen2.5:0.5b ↗494.03M parameters · Q4_K_M292.1 tok/s20,371 tok/sFits0.4 GB0.6 GBOct 21, 2026
lfm2.5-thinking:latest ↗1.2B parameters · Q4_K_M282.2 tok/s10,813 tok/sFits0.7 GB0.9 GBOct 20, 2026
deepseek-coder:latest ↗1B parameters · Q4_0258 tok/s8,205 tok/sFits0.8 GB1.3 GBOct 22, 2026
deepseek-coder:base ↗1B parameters · Q4_0256.8 tok/s7,333 tok/sFits0.8 GB1.3 GBOct 22, 2026
smollm2:360m ↗361.82M parameters · F16252.3 tok/s18,465 tok/sFits0.7 GB0.9 GBOct 22, 2026
qwen3:0.6b ↗751.63M parameters · Q4_K_M247.8 tok/s12,620 tok/sFits0.5 GB0.8 GBOct 21, 2026
granite4:350m ↗352.38M parameters · BF16240.5 tok/s15,228 tok/sFits0.7 GB0.9 GBOct 22, 2026
granite3.1-moe:latest ↗3.3B parameters · Q4_K_M216.3 tok/s7,365 tok/sFits2.0 GB2.2 GBOct 22, 2026
qwen:1.8b-text ↗2B parameters · Q4_0215.3 tok/s7,680 tok/sFits1.1 GB1.5 GBOct 22, 2026
qwen:1.8b ↗2B parameters · Q4_0198.5 tok/s7,649 tok/sFits1.1 GB1.5 GBOct 22, 2026
lfm3-omni:latest ↗1.7B parameters · Q4_K_M198.3 tok/s7,864 tok/sFits2.0 GB2.9 GBOct 31, 2026
lfm3-audio:latest ↗1.7B parameters · Q4_K_M197.7 tok/s7,815 tok/sFits1.7 GB2.2 GBOct 31, 2026
qwen2:1.5b ↗1.5B parameters · Q4_0197.4 tok/s8,656 tok/sFits0.9 GB1.1 GBOct 22, 2026
minicpm-v4.6:latest ↗752.16M parameters · Q4_K_M196 tok/s10,114 tok/sFits1.6 GB1.8 GBOct 20, 2026
qwen3-vl:2b ↗2.1B parameters · Q4_K_M183.2 tok/s7,101 tok/sFits1.9 GB2.8 GBOct 22, 2026
falcon3:3b ↗3.2B parameters · Q4_K_M183 tok/s5,162 tok/sFits2.0 GB2.1 GBOct 22, 2026
deepseek-r1:1.5b ↗1.8B parameters · Q4_K_M182.8 tok/s7,851 tok/sFits1.1 GB1.2 GBOct 21, 2026
llama3.2:1b ↗1.2B parameters · Q8_0182.4 tok/s11,839 tok/sFits1.3 GB1.5 GBOct 21, 2026
qwen3-vl:2b-instruct ↗1.7B parameters · Q4_K_M181.3 tok/s6,989 tok/sFits1.9 GB2.6 GBOct 22, 2026
gemma3:1b ↗999.89M parameters · Q4_K_M180.5 tok/s10,360 tok/sFits0.8 GB1.0 GBOct 21, 2026
qwen2.5:1.5b ↗1.5B parameters · Q4_K_M180.5 tok/s7,927 tok/sFits1.0 GB1.2 GBOct 21, 2026
qwen2.5-coder:1.5b ↗1.5B parameters · Q4_K_M180.4 tok/s7,945 tok/sFits1.0 GB1.2 GBOct 21, 2026
qwen2.5-coder:1.5b-base ↗1.5B parameters · Q4_K_M179.4 tok/s7,867 tok/sFits1.0 GB1.2 GBOct 22, 2026
falcon3:1b ↗1.7B parameters · Q8_0168.6 tok/s10,090 tok/sFits1.8 GB1.7 GBOct 22, 2026
qwen3:1.7b ↗2.0B parameters · Q4_K_M167.6 tok/s7,148 tok/sFits1.4 GB1.5 GBOct 21, 2026
gemma3:1b-it-qat ↗999.89M parameters · Q4_0167.1 tok/s10,858 tok/sFits1.0 GB1.1 GBOct 21, 2026
lfm3:latest ↗1.7B parameters · Q4_K_M154.8 tok/s7,758 tok/sFits1.1 GB1.3 GBOct 31, 2026
lfm3-vl:latest ↗1.7B parameters · Q4_K_M154.6 tok/s7,603 tok/sFits1.5 GB2.0 GBOct 31, 2026
gemma:2b-instruct ↗3B parameters · Q4_0154.2 tok/s7,183 tok/sFits1.6 GB1.7 GBOct 22, 2026
lfm2.5:latest ↗8.5B parameters · Q4_K_M153.4 tok/s5,520 tok/sFits5.2 GB5.0 GBOct 20, 2026
granite4:7b-a1b-h ↗6.9B parameters · Q4_K_M153 tok/s3,207 tok/sFits4.2 GB4.2 GBOct 22, 2026
gemma:2b ↗3B parameters · Q4_0148.1 tok/s7,176 tok/sFits1.7 GB1.8 GBOct 22, 2026
gemma:2b-text ↗3B parameters · Q4_0145.4 tok/s7,110 tok/sFits1.7 GB1.9 GBOct 22, 2026
smollm2:latest ↗1.7B parameters · Q8_0142.1 tok/s6,497 tok/sFits1.8 GB2.3 GBOct 22, 2026
bonsai-2:2b ↗1.9B parameters · Q2_0133.6 tok/s6,244 tok/sFits0.9 GB1.3 GBOct 20, 2026
granite4:1b-h ↗1.5B parameters · Q8_0132.6 tok/s3,331 tok/sFits1.6 GB1.7 GBOct 22, 2026
qwen3.5:0.8b ↗752.39M parameters · Q8_0 · Qwen3.5-0.8B profile132.2 tok/s8,581 tok/sFits1.2 GB1.7 GBOct 20, 2026
gemma4:e2b ↗4.6B parameters · Q4_K_M123.2 tok/s4,296 tok/sFits4.6 GB3.0 GBOct 20, 2026
gemma4:e2b-it-qat ↗4.6B parameters · Q4_0123.2 tok/s4,988 tok/sFits4.3 GB2.7 GBOct 20, 2026
qwen2.5vl:3b ↗3.8B parameters · Q4_K_M119.2 tok/s4,312 tok/sFits3.2 GB3.8 GBOct 22, 2026
gemma2:2b ↗2.6B parameters · Q4_0118.8 tok/s5,439 tok/sFits1.6 GB2.0 GBOct 21, 2026
llama3.2:latest ↗3.2B parameters · Q4_K_M117.7 tok/s4,279 tok/sFits2.0 GB2.3 GBOct 21, 2026
qwen2.5-coder:3b ↗3.1B parameters · Q4_K_M117.4 tok/s4,338 tok/sFits1.9 GB2.1 GBOct 21, 2026
qwen2.5-coder:3b-base ↗3.1B parameters · Q4_K_M116.7 tok/s4,275 tok/sFits1.9 GB2.1 GBOct 22, 2026
qwen2.5:3b ↗3.1B parameters · Q4_K_M116.3 tok/s4,319 tok/sFits1.9 GB2.1 GBOct 21, 2026
granite4:3b-h ↗3.2B parameters · Q4_K_M112.8 tok/s2,401 tok/sFits1.9 GB2.1 GBOct 22, 2026
granite4:micro ↗3.4B parameters · Q4_K_M105.6 tok/s3,660 tok/sFits2.1 GB2.3 GBOct 22, 2026
cogito:3b ↗3.6B parameters · Q4_K_M105.5 tok/s4,231 tok/sFits2.2 GB2.3 GBOct 22, 2026
granite4:latest ↗3.4B parameters · Q4_K_M105.5 tok/s3,612 tok/sFits2.1 GB2.3 GBOct 22, 2026
granite4.1:3b ↗3.4B parameters · Q4_K_M105.2 tok/s3,592 tok/sFits2.1 GB2.3 GBOct 20, 2026
granite4.2:3b ↗3.7B parameters · Q4_K_M105 tok/s3,610 tok/sFits2.2 GB2.3 GBOct 20, 2026
gemma3n:e2b ↗4.5B parameters · Q4_K_M104.9 tok/s2,761 tok/sFits7.8 GB3.0 GBOct 22, 2026
phi3:latest ↗3.8B parameters · Q4_0103.8 tok/s3,657 tok/sFits2.2 GB3.0 GBOct 22, 2026
ministral-3:3b ↗3.8B parameters · Q4_K_M101 tok/s5,385 tok/sFits3.0 GB3.2 GBOct 21, 2026
qwen:4b-text ↗4B parameters · Q4_0100.2 tok/s3,540 tok/sFits2.3 GB3.0 GBOct 22, 2026
qwen:latest ↗4B parameters · Q4_099.6 tok/s3,543 tok/sFits2.3 GB3.0 GBOct 22, 2026
nemotron-3-nano:4b ↗4.0B parameters · Q4_K_M98.8 tok/s2,286 tok/sFits2.8 GB2.8 GBOct 21, 2026
bonsai-2:4b ↗4.2B parameters · Q2_095.8 tok/s2,794 tok/sFits1.6 GB2.0 GBOct 20, 2026
phi3:mini-4k ↗3.8B parameters · Q4_K_M95.7 tok/s3,322 tok/sFits2.4 GB3.2 GBOct 22, 2026
qwen3-vl:4b ↗4.0B parameters · Q4_K_M93 tok/s3,289 tok/sFits3.3 GB4.0 GBOct 22, 2026
qwen3.5:2b ↗1.9B parameters · Q8_093 tok/s6,609 tok/sFits2.7 GB2.9 GBOct 20, 2026
gemma4:latest ↗7.5B parameters · Q4_K_M92 tok/s2,543 tok/sFits6.6 GB4.5 GBOct 20, 2026
qwen3-vl:4b-instruct ↗4.0B parameters · Q4_K_M91.8 tok/s3,300 tok/sFits3.3 GB4.0 GBOct 22, 2026
qwen3:4b-instruct ↗4.0B parameters · Q4_K_M90.5 tok/s3,277 tok/sFits2.5 GB2.8 GBOct 21, 2026
qwen3:4b ↗4.0B parameters · Q4_K_M90 tok/s3,264 tok/sFits2.5 GB2.8 GBOct 21, 2026
gemma3:latest ↗3.9B parameters · Q4_K_M87.2 tok/s3,381 tok/sFits3.4 GB3.6 GBOct 21, 2026
medgemma:latest ↗4.3B parameters · Q4_K_M86.9 tok/s3,401 tok/sFits3.3 GB3.6 GBOct 20, 2026
medgemma1.5:latest ↗4.3B parameters · Q4_K_M86.8 tok/s3,351 tok/sFits3.3 GB3.6 GBOct 20, 2026
translategemma:latest ↗4.3B parameters · Q4_K_M86.6 tok/s3,369 tok/sFits3.3 GB3.5 GBOct 20, 2026
granite4:1b ↗1.6B parameters · BF1683.4 tok/s4,874 tok/sFits3.3 GB3.4 GBOct 22, 2026
gemma3:4b-it-qat ↗3.9B parameters · Q4_076.7 tok/s3,668 tok/sFits4.0 GB4.2 GBOct 21, 2026
deepseek-coder:6.7b-base ↗7B parameters · Q4_074.3 tok/s2,116 tok/sFits3.8 GB4.8 GBOct 22, 2026
bonsai-2:latest ↗9.0B parameters · Q2_074.2 tok/s1,812 tok/sFits3.2 GB3.4 GBOct 20, 2026
gemma4:e4b-it-qat ↗7.5B parameters · Q4_074 tok/s2,944 tok/sFits6.1 GB4.1 GBOct 20, 2026
deepseek-coder:6.7b ↗7B parameters · Q4_073.6 tok/s2,242 tok/sFits3.8 GB4.8 GBOct 22, 2026
codellama:code ↗7B parameters · Q4_073.5 tok/s2,112 tok/sFits3.8 GB4.8 GBOct 22, 2026
llama2:text ↗7B parameters · Q4_073.3 tok/s2,104 tok/sFits3.8 GB4.8 GBOct 22, 2026
codellama:latest ↗7B parameters · Q4_072.9 tok/s2,172 tok/sFits3.8 GB4.8 GBOct 22, 2026
codellama:python ↗7B parameters · Q4_072.9 tok/s2,107 tok/sFits3.8 GB4.8 GBOct 22, 2026
llama2:latest ↗7B parameters · Q4_072.6 tok/s2,164 tok/sFits3.8 GB4.8 GBOct 22, 2026
mistral:v0.1 ↗7B parameters · Q4_071.9 tok/s2,296 tok/sFits4.1 GB4.3 GBOct 22, 2026
mistral:text ↗7B parameters · Q4_071.6 tok/s2,295 tok/sFits4.1 GB4.3 GBOct 22, 2026
mistral:v0.2 ↗7B parameters · Q4_071.5 tok/s2,292 tok/sFits4.1 GB4.3 GBOct 22, 2026
llava:latest ↗7B parameters · Q4_071.4 tok/s2,265 tok/sFits4.7 GB4.9 GBOct 22, 2026
mistral:instruct ↗7.2B parameters · Q4_071.3 tok/s2,284 tok/sFits4.1 GB4.3 GBOct 22, 2026
qwen3.5:4b ↗4.2B parameters · Q4_K_M70.5 tok/s2,858 tok/sFits3.3 GB3.6 GBOct 20, 2026
gemma3n:latest ↗6.9B parameters · Q4_K_M70.4 tok/s1,822 tok/sFits9.8 GB4.2 GBOct 22, 2026
minicpm-v:latest ↗7.6B parameters · Q4_069.2 tok/s2,511 tok/sFits5.5 GB5.2 GBOct 22, 2026
qwen2:7b-text ↗7.6B parameters · Q4_069.2 tok/s2,540 tok/sFits4.4 GB4.2 GBOct 22, 2026
qwen2:latest ↗7.6B parameters · Q4_069.1 tok/s2,528 tok/sFits4.4 GB4.2 GBOct 22, 2026
lfm2:latest ↗23.8B parameters · Q4_K_M67.4 tok/s1,398 tok/sDoesn’t fit · 99% on GPU14.0 GB10.9 GBOct 20, 2026
llama3:text ↗8.0B parameters · Q4_066.9 tok/s2,269 tok/sFits4.7 GB4.6 GBOct 22, 2026
qwen:7b ↗8B parameters · Q4_066.9 tok/s2,119 tok/sFits4.5 GB5.2 GBOct 22, 2026
mistral:latest ↗7.2B parameters · Q4_K_M66.6 tok/s2,066 tok/sFits4.4 GB4.5 GBOct 22, 2026
qwen:7b-text ↗8B parameters · Q4_066.6 tok/s2,101 tok/sFits4.5 GB5.2 GBOct 22, 2026
llama3:latest ↗8.0B parameters · Q4_065.4 tok/s2,264 tok/sFits4.7 GB4.6 GBOct 22, 2026
nemotron-3.5-bolt:e4b ↗7.5B parameters · Q4_K_M64.7 tok/s2,200 tok/sFits7.2 GB6.2 GBOct 26, 2026
qwen2.5vl:latest ↗8.3B parameters · Q4_K_M64.7 tok/s2,293 tok/sFits6.0 GB6.2 GBOct 22, 2026
olmo2:latest ↗7.3B parameters · Q4_K_M64.5 tok/s2,005 tok/sFits4.5 GB5.3 GBOct 22, 2026
qwen2.5-coder:7b-base ↗7.6B parameters · Q4_K_M64 tok/s2,284 tok/sFits4.7 GB4.5 GBOct 22, 2026
falcon3:latest ↗7.5B parameters · Q4_K_M63.9 tok/s2,181 tok/sFits4.6 GB4.6 GBOct 22, 2026
olmo-3:7b-instruct ↗7.3B parameters · Q4_K_M63.7 tok/s1,735 tok/sFits4.5 GB5.3 GBOct 21, 2026
qwen2.5:latest ↗7.6B parameters · Q4_K_M63.7 tok/s2,295 tok/sFits4.7 GB4.5 GBOct 21, 2026
olmo-3:latest ↗7.3B parameters · Q4_K_M63.6 tok/s1,742 tok/sFits4.5 GB5.3 GBOct 21, 2026
deepseek-r1:7b ↗7.6B parameters · Q4_K_M63.5 tok/s2,274 tok/sFits4.7 GB4.5 GBOct 21, 2026
qwen2.5-coder:latest ↗7.6B parameters · Q4_K_M63.4 tok/s2,292 tok/sFits4.7 GB4.5 GBOct 21, 2026
cogito:latest ↗8.0B parameters · Q4_K_M61.9 tok/s2,063 tok/sFits4.9 GB4.8 GBOct 22, 2026
dolphin3:latest ↗8.0B parameters · Q4_K_M61.7 tok/s2,090 tok/sFits4.9 GB4.8 GBOct 22, 2026
llama3.1:latest ↗8.0B parameters · Q4_K_M61.4 tok/s2,066 tok/sFits4.9 GB4.8 GBOct 21, 2026
qwen3-vl:8b-instruct ↗8.2B parameters · Q4_K_M59.9 tok/s1,974 tok/sFits6.2 GB6.4 GBOct 22, 2026
qwen3-vl:latest ↗8.2B parameters · Q4_K_M59.9 tok/s1,963 tok/sFits6.2 GB6.4 GBOct 22, 2026
minicpm-v4.5:latest ↗8.2B parameters · Q4_K_M59.1 tok/s1,968 tok/sFits6.1 GB6.0 GBOct 20, 2026
lfm2.5:24b ↗24.0B parameters · Q4_K_M58 tok/s1,108 tok/sDoesn’t fit · 99% on GPU15.0 GB10.9 GBOct 20, 2026
deepseek-r1:latest ↗8.2B parameters · Q4_K_M57.7 tok/s1,983 tok/sFits5.2 GB5.1 GBOct 21, 2026
qwen3:latest ↗8.2B parameters · Q4_K_M57.1 tok/s1,991 tok/sFits5.2 GB5.1 GBOct 21, 2026
gemma:latest ↗9B parameters · Q4_056.8 tok/s1,917 tok/sFits5.0 GB5.9 GBOct 22, 2026
gemma:text ↗9B parameters · Q4_055.1 tok/s1,868 tok/sFits5.2 GB6.1 GBOct 22, 2026
granite4.1:8b ↗8.8B parameters · Q4_K_M54.8 tok/s1,727 tok/sFits5.3 GB5.3 GBOct 20, 2026
granite4.2:latest ↗8.8B parameters · Q4_K_M54.8 tok/s1,716 tok/sFits5.3 GB5.3 GBOct 20, 2026
ornith:latest ↗9.0B parameters · Q4_K_M54.2 tok/s1,820 tok/sFits5.6 GB5.1 GBOct 20, 2026
ornith-1.5:9b ↗9.0B parameters · Q4_K_M54.1 tok/s1,789 tok/sFits6.6 GB6.2 GBOct 20, 2026
ministral-3:latest ↗8.9B parameters · Q4_K_M53.5 tok/s2,567 tok/sFits6.0 GB5.9 GBOct 21, 2026
gemma4:12b ↗11.9B parameters · Q4_K_M53.1 tok/s1,014 tok/sFits7.7 GB8.3 GBOct 20, 2026
gemma2:latest ↗9.2B parameters · Q4_050.1 tok/s1,705 tok/sFits5.4 GB6.0 GBOct 21, 2026
qwen3.5:latest ↗9.0B parameters · Q4_K_M47.9 tok/s1,794 tok/sFits6.6 GB6.2 GBOct 20, 2026
falcon3:10b ↗10.3B parameters · Q4_K_M46.5 tok/s1,518 tok/sFits6.3 GB6.3 GBOct 22, 2026
laguna-xs-2.1:latest ↗33.4B parameters · Q4_K_M45.4 tok/s518 tok/sDoesn’t fit · 98% on GPU20.0 GB10.8 GBOct 20, 2026
mistral-nemo:latest ↗12.2B parameters · Q4_044.2 tok/s1,509 tok/sFits7.1 GB6.8 GBOct 22, 2026
ornith:35b ↗34.7B parameters · Q4_K_M41.5 tok/s452 tok/sDoesn’t fit · 98% on GPU21.0 GB10.7 GBOct 20, 2026
codellama:13b-code ↗13B parameters · Q4_041 tok/s1,181 tok/sFits7.4 GB8.7 GBOct 22, 2026
llama2:13b-text ↗13B parameters · Q4_041 tok/s1,189 tok/sFits7.4 GB8.7 GBOct 22, 2026
llava:13b ↗13B parameters · Q4_040.5 tok/s1,199 tok/sFits8.0 GB9.3 GBOct 22, 2026
codellama:13b ↗13B parameters · Q4_040.4 tok/s1,203 tok/sFits7.4 GB8.7 GBOct 22, 2026
codellama:13b-python ↗13B parameters · Q4_040.4 tok/s1,181 tok/sFits7.4 GB8.7 GBOct 22, 2026
llama2:13b ↗13B parameters · Q4_040.3 tok/s1,207 tok/sFits7.4 GB8.7 GBOct 22, 2026
gemma4:12b-it-qat ↗11.9B parameters · Q4_039.4 tok/s1,259 tok/sFits7.2 GB7.4 GBOct 20, 2026
gpt-oss:latest ↗20.9B parameters · MXFP439.4 tok/s1,535 tok/sDoesn’t fit · 90% on GPU14.0 GB10.1 GBOct 21, 2026
flint:latest ↗39.8B parameters · NVFP439.2 tok/s437 tok/sDoesn’t fit · 91% on GPU23.0 GB10.1 GBOct 26, 2026
phi3:medium ↗14.0B parameters · Q4_039.2 tok/s1,245 tok/sFits7.9 GB8.0 GBOct 22, 2026
phi3:medium-4k ↗14.0B parameters · Q4_039.2 tok/s1,248 tok/sFits7.9 GB8.0 GBOct 22, 2026
gemma3:12b ↗11.8B parameters · Q4_K_M38 tok/s1,175 tok/sFits8.2 GB8.4 GBOct 21, 2026
translategemma:12b ↗12.2B parameters · Q4_K_M37.9 tok/s1,178 tok/sFits8.1 GB8.3 GBOct 20, 2026
qwen:14b ↗14B parameters · Q4_037.8 tok/s1,170 tok/sFits8.2 GB9.1 GBOct 22, 2026
qwen:14b-text ↗14B parameters · Q4_037.7 tok/s1,168 tok/sFits8.2 GB9.1 GBOct 22, 2026
qwen3.5:35b ↗34.7B parameters · Q4_K_M37.3 tok/s431 tok/sDoesn’t fit · 98% on GPU22.0 GB11.5 GBOct 20, 2026
olmo2:13b ↗13.7B parameters · Q4_K_M36 tok/s1,103 tok/sFits8.4 GB9.4 GBOct 22, 2026
gemma3:12b-it-qat ↗11.8B parameters · Q4_035.6 tok/s1,281 tok/sFits8.9 GB9.1 GBOct 21, 2026
gemma4:26b-a4b-it-qat ↗25.2B parameters · Q4_0 · gemma-4-26B-A4B-it-qat-q4_0 profile35.2 tok/s598 tok/sDoesn’t fit · 96% on GPU16.0 GB9.6 GBOct 20, 2026
ministral-3:14b ↗13.9B parameters · Q4_K_M35 tok/s1,656 tok/sFits9.1 GB8.8 GBOct 21, 2026
phi4:latest ↗14.7B parameters · Q4_K_M34.7 tok/s1,119 tok/sFits9.1 GB8.9 GBOct 22, 2026
cogito:14b ↗14.8B parameters · Q4_K_M33.8 tok/s1,065 tok/sFits9.0 GB8.6 GBOct 22, 2026
north-mini-code-1.0:latest ↗30.5B parameters · Q4_K_M33.8 tok/s547 tok/sDoesn’t fit · 99% on GPU19.0 GB10.9 GBOct 20, 2026
qwen3:14b ↗14.8B parameters · Q4_K_M33.6 tok/s1,105 tok/sFits9.3 GB8.9 GBOct 21, 2026
qwen2.5-coder:14b ↗14.8B parameters · Q4_K_M33.5 tok/s1,080 tok/sFits9.0 GB8.6 GBOct 21, 2026
deepseek-r1:14b ↗14.8B parameters · Q4_K_M33.4 tok/s1,062 tok/sFits9.0 GB8.6 GBOct 21, 2026
qwen2.5-coder:14b-base ↗14.8B parameters · Q4_K_M33.4 tok/s1,072 tok/sFits9.0 GB8.6 GBOct 22, 2026
qwen2.5:14b ↗14.8B parameters · Q4_K_M33.3 tok/s1,081 tok/sFits9.0 GB8.6 GBOct 21, 2026
qwen3:30b ↗30.5B parameters · Q4_K_M33.1 tok/s731 tok/sDoesn’t fit · 98% on GPU19.0 GB10.1 GBOct 21, 2026
qwen3:30b-instruct ↗30.5B parameters · Q4_K_M33.1 tok/s737 tok/sDoesn’t fit · 98% on GPU19.0 GB10.1 GBOct 21, 2026
qwen3-vl:30b ↗31.1B parameters · Q4_K_M32.9 tok/s566 tok/sDoesn’t fit · 98% on GPU20.0 GB11.4 GBOct 22, 2026
qwen3-vl:30b-a3b-instruct ↗30.5B parameters · Q4_K_M32.7 tok/s555 tok/sDoesn’t fit · 98% on GPU20.0 GB11.5 GBOct 22, 2026
bonsai-2:27b ↗26.9B parameters · Q2_032.5 tok/s583 tok/sFits8.3 GB8.2 GBOct 20, 2026
qwen3-coder:latest ↗30.5B parameters · Q4_K_M31.5 tok/s720 tok/sDoesn’t fit · 98% on GPU19.0 GB10.1 GBOct 21, 2026
glm-4.7-flash:latest ↗29.9B parameters · Q4_K_M31 tok/s473 tok/sDoesn’t fit · 98% on GPU19.0 GB10.8 GBOct 20, 2026
laguna-xs.2:latest ↗33.4B parameters · Q4_K_M30.5 tok/s437 tok/sDoesn’t fit · 96% on GPU23.0 GB10.8 GBOct 20, 2026
gemma4:26b ↗25.2B parameters · Q4_K_M29.6 tok/s503 tok/sDoesn’t fit · 68% on GPU16.0 GB10.2 GBOct 20, 2026
nemotron-cascade-2:latest ↗31.6B parameters · Q4_K_M28.1 tok/s443 tok/sDoesn’t fit · 98% on GPU18.0 GB10.7 GBOct 20, 2026
nemotron-3-nano:latest ↗31.6B parameters · Q4_K_M27 tok/s437 tok/sDoesn’t fit · 98% on GPU18.0 GB10.5 GBOct 21, 2026
nemotron3:latest ↗33.0B parameters · Q4_K_M25.5 tok/s363 tok/sDoesn’t fit · 97% on GPU21.0 GB9.3 GBOct 20, 2026
qwen3.6:35b-a3b-coding ↗35.5B parameters · Q4_K_M18.6 tok/s410 tok/sDoesn’t fit · 96% on GPU23.0 GB9.8 GBOct 20, 2026
nemotron-3.5-lightning:latest ↗32.9B parameters · Q4_K_M18.2 tok/s418 tok/sDoesn’t fit · 96% on GPU19.0 GB10.6 GBOct 20, 2026
qwen3.6:latest ↗35.5B parameters · Q4_K_M17.3 tok/s423 tok/sDoesn’t fit · 98% on GPU23.0 GB11.2 GBOct 30, 2026
phi4-reasoning:latest ↗14.7B parameters · Q4_K_M14.9 tok/s1,172 tok/sDoesn’t fit · 90% on GPU11.0 GB10.1 GBOct 22, 2026
phi4-reasoning:plus ↗14.7B parameters · Q4_K_M14.9 tok/s1,155 tok/sDoesn’t fit · 90% on GPU11.0 GB10.1 GBOct 22, 2026
granite4:32b-a9b-h ↗32.2B parameters · Q4_K_M14 tok/s364 tok/sDoesn’t fit · 98% on GPU19.0 GB10.0 GBOct 22, 2026
mistral-small:22b ↗22.2B parameters · Q4_07.5 tok/s505 tok/sDoesn’t fit · 91% on GPU13.0 GB10.1 GBOct 22, 2026
mistral-small:latest ↗23.6B parameters · Q4_K_M5.4 tok/s471 tok/sDoesn’t fit · 72% on GPU14.0 GB10.2 GBOct 22, 2026
mistral-small3.2:latest ↗24.0B parameters · Q4_K_M5.3 tok/s575 tok/sDoesn’t fit · 95% on GPU15.0 GB11.0 GBOct 22, 2026
qwen3.6:27b ↗27.3B parameters · Q4_K_M5.1 tok/s239 tok/sDoesn’t fit · 60% on GPU18.0 GB11.4 GBOct 20, 2026
qwen3.6:27b-coding ↗27.3B parameters · Q4_K_M5.1 tok/s238 tok/sDoesn’t fit · 60% on GPU18.0 GB11.4 GBOct 20, 2026
qwen3.8:latest ↗27.3B parameters · Q4_K_M4.1 tok/s222 tok/sDoesn’t fit · 59% on GPU18.0 GB11.2 GBOct 20, 2026
qwen3.5:27b ↗26.9B parameters · Q4_K_M3.9 tok/s300 tok/sDoesn’t fit · 66% on GPU17.0 GB11.5 GBOct 20, 2026
gemma4:31b ↗30.7B parameters · Q4_K_M3.8 tok/s183 tok/sFits19.0 GB10.1 GBOct 20, 2026
gemma2:27b ↗27.2B parameters · Q4_03.7 tok/s295 tok/sDoesn’t fit · 64% on GPU16.0 GB10.1 GBOct 21, 2026
translategemma:27b ↗27.4B parameters · Q4_K_M3.7 tok/s278 tok/sDoesn’t fit · 89% on GPU17.0 GB11.6 GBOct 20, 2026
muse-glimmer:latest ↗27.9B parameters · Q4_K_M3.4 tok/s286 tok/sDoesn’t fit · 57% on GPU18.0 GB9.4 GBOct 20, 2026
gemma3:27b ↗27.0B parameters · Q4_K_M3.3 tok/s262 tok/sDoesn’t fit · 61% on GPU17.0 GB10.9 GBOct 21, 2026
medgemma:27b ↗27.4B parameters · Q4_K_M3.3 tok/s260 tok/sDoesn’t fit · 89% on GPU17.0 GB10.9 GBOct 21, 2026
granite4.1:30b ↗28.9B parameters · Q4_K_M3.2 tok/s274 tok/sDoesn’t fit · 62% on GPU17.0 GB10.8 GBOct 20, 2026
granite4.2:30b ↗29.3B parameters · Q4_K_M3.2 tok/s281 tok/sDoesn’t fit · 61% on GPU18.0 GB10.8 GBOct 20, 2026
gemma3:27b-it-qat ↗27.0B parameters · Q4_03.1 tok/s264 tok/sDoesn’t fit · 60% on GPU18.0 GB11.1 GBOct 21, 2026
qwen:32b ↗33B parameters · Q4_02.8 tok/s254 tok/sDoesn’t fit · 55% on GPU18.0 GB10.2 GBOct 22, 2026
qwen:32b-text ↗33B parameters · Q4_02.8 tok/s262 tok/sDoesn’t fit · 55% on GPU18.0 GB10.2 GBOct 22, 2026
codellama:34b-python ↗34B parameters · Q4_02.6 tok/s240 tok/sDoesn’t fit · 53% on GPU19.0 GB9.9 GBOct 22, 2026
gemma4:31b-it-qat ↗30.7B parameters · Q4_02.6 tok/s216 tok/sDoesn’t fit · 95% on GPU19.0 GB9.5 GBOct 20, 2026
codellama:34b ↗34B parameters · Q4_02.5 tok/s245 tok/sDoesn’t fit · 53% on GPU19.0 GB9.9 GBOct 22, 2026
deepseek-coder:33b ↗33B parameters · Q4_02.5 tok/s255 tok/sDoesn’t fit · 54% on GPU19.0 GB10.0 GBOct 22, 2026
olmo-3:32b ↗32.2B parameters · Q4_K_M2.5 tok/s203 tok/sDoesn’t fit · 95% on GPU19.0 GB10.1 GBOct 21, 2026
olmo-3.1:32b-instruct ↗32.2B parameters · Q4_K_M2.5 tok/s209 tok/sDoesn’t fit · 95% on GPU19.0 GB10.1 GBOct 21, 2026
olmo-3.1:latest ↗32.2B parameters · Q4_K_M2.5 tok/s208 tok/sDoesn’t fit · 95% on GPU19.0 GB10.1 GBOct 21, 2026
cogito:32b ↗32.8B parameters · Q4_K_M2.4 tok/s228 tok/sDoesn’t fit · 51% on GPU20.0 GB10.0 GBOct 22, 2026
llava:34b ↗34B parameters · Q4_02.4 tok/s244 tok/sDoesn’t fit · 52% on GPU20.0 GB10.8 GBOct 22, 2026
qwen2.5-coder:32b ↗32.8B parameters · Q4_K_M2.4 tok/s226 tok/sDoesn’t fit · 51% on GPU20.0 GB10.0 GBOct 21, 2026
qwen2.5-coder:32b-base ↗32.8B parameters · Q4_K_M2.4 tok/s222 tok/sDoesn’t fit · 51% on GPU20.0 GB10.0 GBOct 22, 2026
qwen2.5:32b ↗32.8B parameters · Q4_K_M2.4 tok/s227 tok/sDoesn’t fit · 51% on GPU20.0 GB10.0 GBOct 21, 2026
qwen2.5vl:32b ↗33.5B parameters · Q4_K_M2.4 tok/s226 tok/sDoesn’t fit · 93% on GPU21.0 GB11.2 GBOct 22, 2026
qwen3-vl:32b ↗33.4B parameters · Q4_K_M2.4 tok/s225 tok/sDoesn’t fit · 93% on GPU21.0 GB11.5 GBOct 22, 2026
qwen3-vl:32b-instruct ↗33.4B parameters · Q4_K_M2.4 tok/s226 tok/sDoesn’t fit · 93% on GPU21.0 GB11.5 GBOct 22, 2026
deepseek-r1:32b ↗32.8B parameters · Q4_K_M2.3 tok/s224 tok/sDoesn’t fit · 51% on GPU20.0 GB10.0 GBOct 21, 2026
qwen3:32b ↗32.8B parameters · Q4_K_M2.3 tok/s222 tok/sDoesn’t fit · 51% on GPU20.0 GB10.1 GBOct 21, 2026
20 models couldn’t run on this machine

Failed tests are tried again after 14 days, sooner when Ollama is updated.

How these speeds are measured

Each model is loaded once to warm up, then asked to write 512 tokens 3 times. Writing speed is the median of those runs. Reading speed comes from one prompt of about 1,500 tokens. All runs use temperature 0, a 4,096-token context, reasoning (“thinking”) turned off where the model supports it, and Ollama’s default settings and download for each tag, with nothing else using the graphics card. Every morning the lab tests new models first, then re-times its oldest results so they keep up with Ollama and driver updates.

“Fits in 12 GB” means Ollama kept the whole model in the graphics card’s memory. “Doesn’t fit” shows the share that did; the rest ran from system memory, which is much slower and depends on this machine’s processor and RAM.

What these numbers don’t tell you

Speed is not quality: a fast model can give worse answers. Results change with the runtime version, driver, context length, longer conversations and running several requests at once. Downloads larger than 23 GB and models that can’t write text are not tested.

Download and memory figures use decimal GB. See models that should fit a 12GB budget · Read the memory guide · Compare ratings