LLMs · Meta

Llama 3.2 90B Vision Instruct

Name: Llama 3.2 90B Vision Instruct
Author: Meta

Parameters: 89B
Architecture: Dense
Max context: 128K tokens
License: llama3.2
Released: 2024-09

How to run it

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

ollama run llama3.2-vision:90b

Original weights · Hugging Face

Full-precision files from Meta, for fine-tuning, vLLM, Transformers.

huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct (opens in a new tab)

Free Hugging Face account needed: request access on the model page, and Meta approves it (usually within minutes to a few hours).

Links checked 3 days ago against Hugging Face, Ollama and GitHub.

Apps that run Llama 3.2 90B Vision Instruct

Our pick for each kind of computer, assuming the model fits in its memory. Pick your own GPU or device in the Model Finder for an exact answer.

Mac (Apple silicon)

LM Studio

Runs the MLX build, typically up to 10% faster than GGUF for a model with 88.6B active parameters on Apple silicon.

Also good:

Windows PC, NVIDIA GPU

Ollama

One command with Ollama’s own build of this model.

Also good:

Windows PC, AMD Radeon

Ollama

One command with Ollama’s own build of this model.

Also good:

Linux, NVIDIA GPU

Ollama

One command with Ollama’s own build of this model.

Also good:

Linux server, 80 GB+ GPUs

vLLM

The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.

Also good:

App	Type	Opens it as	Systems	Engine	Start
LM Studio	Desktop app	MLX	macOSWindowsLinux	llama.cpp, MLX	Open in LM Studio
Ollama	CLI + API	Ollama library	macOSWindowsLinux	llama.cpp, MLX	`ollama run llama3.2-vision:90b`
Jan	Desktop app	MLX	macOSWindowsLinux	llama.cpp, MLX	Get Jan
MLX LM	CLI + API	MLX	macOS	MLX	`mlx_lm.chat --model mlx-community/Llama-3.2-90B-Vision-Instruct-4bit`
vLLM	Server	Original weights	WSLWindowsLinux	vLLM	`vllm serve meta-llama/Llama-3.2-90B-Vision-Instruct`
SGLang	Server	Original weights	WSLWindowsLinux	SGLang	`python3 -m sglang.launch_server --model-path meta-llama/Llama-3.2-90B-Vision-Instruct`
Docker Model Runner	CLI + API	Original weights	macOSWindowsLinux	llama.cpp, vLLM	`docker model run hf.co/meta-llama/Llama-3.2-90B-Vision-Instruct`
Unsloth Desktop	Desktop app	MLX	macOSWindowsLinux	llama.cpp, MLX, stable-diffusion.cpp	Get Unsloth Desktop
Atomic Chat	Desktop app	MLX	macOSWindowsLinux	llama.cpp, MLX-VLM	Open in Atomic Chat
Text Generation WebUI	Desktop app	Original weights	macOSWindowsLinux	llama.cpp, Transformers, ExLlamaV3	Get Text Generation WebUI
Msty Studio	Desktop app	MLX	macOSWindowsLinux	Ollama, llama.cpp, MLX	Get Msty Studio
Inferencer	Desktop app	MLX	macOS	MLX	Get Inferencer
exo	Server	MLX	macOSLinux	MLX	Get exo

Which apps can open a model depends on the files published for it: GGUF for llama.cpp-based apps, MLX for Apple’s engine, the original weights for vLLM and SGLang. Compare every app.

Memory needed

Weights plus KV cache and runtime overhead. Leave a little extra free for the operating system and other apps. Try another context length.

Quantization	Quality	Weights	Total at 8K context	Total at 32K context	Download
`Q2_K`	Low	35.4 GB	39.4 GB	49.5 GB	All files (opens in a new tab)
`Q3_K_M`	Low	43.4 GB	47.4 GB	57.4 GB	All files (opens in a new tab)
`Q4_K_M`	Good	54.0 GB	58.0 GB	68.1 GB
`Q5_K_M`	High	62.9 GB	66.9 GB	76.9 GB	All files (opens in a new tab)
`Q6_K`	High	72.6 GB	76.6 GB	86.7 GB	All files (opens in a new tab)
`Q8_0`	Very high	93.9 GB	97.9 GB	108 GB
`F16`	Full	177 GB	181 GB	191 GB	Original weights (opens in a new tab)

Best hardware for Llama 3.2 90B Vision Instruct

The least expensive GPU in our catalog that runs it without offloading is the AMD Radeon PRO W7800 48GB (launch price $2,799) at Q2_K.

Quality	Hardware	Quant	Memory	Speed	Runs
62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA RTX PRO 6000 Blackwell	`Q4_K_M`	62.8 GB / 96.0 GB	24.8 tok/s	Runs well
55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking.	Apple M2 Ultra (192 GB)	`Q3_K_M`	58.9 GB / 144 GB	12.3 tok/s	Runs well
46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking.	Apple M4 Max (64 GB)	`Q2_K`	43.2 GB / 48.0 GB	10.7 tok/s	Runs welltight fit
55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking.	Apple M3 Ultra (256 GB)	`Q3_K_M`	62.7 GB / 192 GB	12.6 tok/s	Runs well
46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking.	Apple M4 Max (128 GB)	`Q2_K`	47.1 GB / 96.0 GB	10.7 tok/s	Runs well
55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking.	Apple M3 Ultra (512 GB)	`Q3_K_M`	78.1 GB / 384 GB	12.6 tok/s	Runs well
65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA GH200 Grace Hopper (96 GB GPU)	`Q6_K`	81.4 GB / 96.0 GB	38.5 tok/s	Runs great
65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H20	`Q6_K`	81.4 GB / 96.0 GB	38.5 tok/s	Runs great
64Good quality (B) · 64 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA H100 NVL	`Q5_K_M`	71.6 GB / 94.0 GB	44.6 tok/s	Runs great
66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking.	NVIDIA B200	`Q8_0`	107 GB / 180 GB	56.4 tok/s	Runs great

Check it on your own hardware

Run it on a cluster

Devices that can’t run Llama 3.2 90B Vision Instruct alone can run it together. These example clusters do:

Try it with your own devices

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.