# Llama 3.2 90B Vision Instruct ![Organization](https://img.shields.io/badge/Org-Meta-0064e0) ![Parameters](https://img.shields.io/badge/Params-90B-1f6feb) ![Context](https://img.shields.io/badge/Context-128K%20tokens-1f6feb) ![License](https://img.shields.io/badge/License-Llama%203.2-d29922) ![Weights](https://img.shields.io/badge/Weights-Open-2da44e) ![Multimodal](https://img.shields.io/badge/Vision-Yes-2da44e) ![Released](https://img.shields.io/badge/Released-Sep%202024-555) > **Quick answer:** Llama 3.2 90B Vision Instruct is Meta's September 2024 open-weight multimodal model, scoring 68.0% on MATH and 86.0% on MMLU. With a 128K context window and vision capability, it was the first large open-weight model to support image understanding at frontier text-model scale. ## At a Glance **Where Llama 3.2 90B leads** - First large open-weight multimodal model at frontier scale - 86.0% MMLU — strong general knowledge - 128K context window - Vision capability: understands images alongside text - Llama 3.2 Community License (free for most commercial uses) **Where it lags** - 68.0% MATH — moderate mathematics - December 2023 knowledge cutoff — earlier than Llama 3.3 70B - Superseded by Llama 4 Maverick for multimodal tasks (1M context, better benchmarks) - 90B parameters — heavier than Llama 3.3 70B for text-only tasks **Best for:** Open-weight vision-language tasks where Llama 4 Maverick is too recent for existing integrations; image analysis pipelines on Llama 3.2 stack. ## What Llama 3.2 90B Vision Instruct Is Llama 3.2 90B Vision Instruct (released September 25, 2024) was Meta's first large multimodal model — adding image understanding to the flagship Llama scale. Alongside the text-only Llama 3.2 3B and 1B models, it marked Meta's push into multimodal open-weight AI. At 90B parameters, it provides text and vision capability at frontier scale — competitive with GPT-4V and Claude's vision capabilities at launch. The model handles image captioning, visual Q&A, document understanding, and chart analysis alongside standard text tasks. For new multimodal open-weight deployments, Llama 4 Maverick (MoE, 1M context, better benchmarks) is the recommended choice. Llama 3.2 90B remains relevant for teams with existing 3.2 integrations. ## Specifications | Field | Value | |---|---| | Organization | Meta | | Parameters | 90B (dense) | | Context window | 128,000 tokens | | License | Llama 3.2 Community License | | HuggingFace | [meta-llama/Llama-3.2-90B-Vision-Instruct](https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct) | | Release date | September 25, 2024 | | Knowledge cutoff | December 2023 | | Modality | Text + Vision (multimodal) | ## Pricing Open weights under Llama 3.2 Community License. Available via Meta AI API and third-party providers. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [MATH](/benchmarks/math) | 68.0% | Benchgen evaluation | 2025-07 | | [MMLU](/benchmarks/mmlu) | 86.0% | Benchgen evaluation | 2025-07 | ## Llama 3.2 90B vs Alternatives | Model | MMLU | Vision | Context | License | |---|---|---|---|---| | **Llama 3.2 90B Vision Instruct** | 86.0% | Yes | 128K | Llama 3.2 | | [Llama 4 Maverick](/models/meta/llama-4-maverick) | — | Yes | 1M | Llama 4 | | [Gemma 3 27B](/models/google/gemma-3-27b) | — | Yes | 128K | Gemma ToU | Llama 4 Maverick: better benchmarks, 1M context, multimodal — preferred for new deployments. Llama 3.2 90B for existing integrations or when Llama 3.2 stack is required. ## Related Models & Benchmarks - Compare: [Llama 4 Maverick](/models/meta/llama-4-maverick) · [Llama 3.3 70B Instruct](/models/meta/llama-3-3-70b-instruct) · [Gemma 3 27B](/models/google/gemma-3-27b) - Benchmarks: [MATH](/benchmarks/math) · [MMLU](/benchmarks/mmlu) ## Frequently Asked Questions

What is Llama 3.2 90B Vision Instruct?

Meta's September 2024 open-weight multimodal 90B model scoring 68.0% MATH and 86.0% MMLU with 128K context. First large open-weight model with vision capability.

Does Llama 3.2 90B support images?

Yes. Llama 3.2 90B Vision Instruct supports text and image inputs — vision-language tasks including image captioning, visual Q&A, and document understanding.

--- *Specs from Meta's Llama 3.2 announcement (September 2024) and Benchgen evaluations. Last updated 2026-07-24.*

Llama 3.2 90B Instruct — Evaluations

Evaluations