FITMYLLM · AUGUST 17, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K
Quant Bits VRAM @ 16K Quality IQ2_M 2.93 37.2 GB
33.5 + 3.8 KV
low Q2_K 3.16 39.8 GB
36.0 + 3.8 KV
low IQ3_XXS 3.25 40.8 GB
37.1 + 3.8 KV
low IQ3_XS 3.5 43.6 GB
39.9 + 3.8 KV
low Q3_K_S 3.64 45.2 GB
41.4 + 3.8 KV
low IQ3_M 3.76 46.5 GB
42.8 + 3.8 KV
low Q3_K_M 4 49.2 GB
45.5 + 3.8 KV
low Q3_K_L 4.3 52.6 GB
48.9 + 3.8 KV
moderate IQ4_XS 4.46 54.4 GB
50.7 + 3.8 KV
moderate Q4_K_S 4.67 56.8 GB
53.0 + 3.8 KV
moderate Q4_K_M 4.89 59.3 GB
55.5 + 3.8 KV
good Q5_K_S 5.57 66.9 GB
63.2 + 3.8 KV
good Q5_K_M 5.7 68.4 GB
64.6 + 3.8 KV
good Q6_K 6.56 78.0 GB
74.3 + 3.8 KV
excellent Q8_0 8.5 99.9 GB
96.1 + 3.8 KV
lossless FP16 16 184.2 GB
180.5 + 3.8 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Deploying for a team or in production? Size GPUs, cost & scaling in Enterprise → ▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN LLAMA-3.2-90B-VISION-INSTRUCT NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_M — 33.5 GB VRAM Q2_K — 36.0 GB VRAM IQ3_XXS — 37.1 GB VRAM IQ3_XS — 39.9 GB VRAM Q3_K_S — 41.4 GB VRAM IQ3_M — 42.8 GB VRAM Q3_K_M — 45.5 GB VRAM Q3_K_L — 48.9 GB VRAM IQ4_XS — 50.7 GB VRAM Q4_K_S — 53.0 GB VRAM Q4_K_M — 55.5 GB VRAM Q5_K_S — 63.2 GB VRAM Q5_K_M — 64.6 GB VRAM Q6_K — 74.3 GB VRAM Q8_0 — 96.1 GB VRAM FP16 — 180.5 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run llama3.2-vision:90b-instruct-q4_K_MCOPY
Downloads and runs automatically. Add --verbose for speed stats.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for Llama-3.2-90B-Vision-Instruct
Build Hardware for Llama-3.2-90B-Vision-Instruct Llama 3.2 90B Vision Instruct — top open multimodal model for complex visual tasks.
Read full model card ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Llama-3.2-90B-Vision-Instruct — 90B Dense. ▸ SPECIFICATIONS
PARAMETERS 90B
ARCHITECTURE Dense Transformer
CONTEXT LENGTH 128K tokens
CAPABILITIES vision, chat
RELEASE DATE 2024-09-25
PROVIDER Meta
FAMILY llama ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_M 2.93 33.5 GB 75% Q2_K 3.16 36.0 GB 78% IQ3_XXS 3.25 37.1 GB 82% IQ3_XS 3.5 39.9 GB 84% Q3_K_S 3.64 41.4 GB 85% IQ3_M 3.76 42.8 GB 86% Q3_K_M 4 45.5 GB 88% Q3_K_L 4.3 48.9 GB 90% IQ4_XS 4.46 50.7 GB 92% Q4_K_S 4.67 53.0 GB 93% Q4_K_M 4.89 55.5 GB 94% Q5_K_S 5.57 63.2 GB 96% Q5_K_M 5.7 64.6 GB 96% Q6_K 6.56 74.3 GB 97% Q8_0 8.5 96.1 GB 100% FP16 16 180.5 GB 100%
§ 01 BENCHMARK SCORES
HumanEval 65.0
MMLU-PRO 68.0
MATH 52.0
IFEval 82.0
BBH 70.0
MMMU 60.3
GPQA 48.0
MUSR 25.0
MMBench 85.5
LiveCodeBench 21.4
AIME 5.0
GPQA Diamond 43.2
HLE 4.9
§ 02 RUN COMMAND
Run Llama-3.2-90B-Vision-Instruct locally with Ollama — needs 55.5 GB VRAM at Q4_K_M:
$ ollama run llama3.2-vision:90b
§ 03 COMPATIBLE GPUs
30 @ Q4_K_M Feedback 