BestLLMfor EN Your hardware. Your LLM. Your call.
APIOpen data Find my LLM
Model fiche

LLaVA-OneVision 72B

By LMMs-Lab · Singapore

Updated 2026-07-13

vision chat
Parameters
72B
License
Apache 2.0
Context
32k
VRAM (Q4)
42 GB
Released
September 2024

Overview

The 72B Apache-licensed flagship from LMMs-Lab, built on Qwen2-72B with strong English and Chinese vision performance. A 2024 state-of-the-art open VLM.

When to pick this model

  • High-fidelity image and video understanding on owned hardware
  • Commercial deployments needing Apache-licensed VLMs at scale
  • Multi-image reasoning over technical documents
  • Bilingual EN/CN visual tasks

VRAM requirements by quantization

VRAM REQUIRED (GB)121624324880128Q4_K_M42 GBQ5_K_M50 GBQ8_078 GBFP16144 GB
QuantizationVRAM required
Q4_K_M (recommended)42 GB
Q5_K_M50 GB
Q8_078 GB
FP16 (no quantization)144 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

In practice, LLaVA-OneVision 72B spills past single consumer GPUs even at Q4_K_M (42 GB) — think dual-GPU or workstation cards. Stepping up to Q8_0 nearly doubles the footprint to 78 GB, and unquantized FP16 weights take 144 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, LLaVA-OneVision 72B needs roughly 64 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 1 tokens/sec on entry-level GPUs, on the order of 5 tokens/sec on a mid-range card, and up to 18 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.

What hardware do you need

The table below matches LLaVA-OneVision 72B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.

GPU memoryExample cardsBest fit for LLaVA-OneVision 72B
8 GBRTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GBDoes not fit — needs 42 GB at Q4_K_M
12 GBRTX 5070, RTX 5070 Ti Laptop, RTX 4080 LaptopDoes not fit — needs 42 GB at Q4_K_M
16 GBRTX 5080, RTX 4080 Super, Radeon RX 9070 XTDoes not fit — needs 42 GB at Q4_K_M
24 GBRTX 4090, Radeon RX 7900 XTX, RTX 5090 LaptopDoes not fit — needs 42 GB at Q4_K_M
32 GBRTX 5090Does not fit — needs 42 GB at Q4_K_M

Which GPU should you buy to run LLaVA-OneVision 72B?

To run LLaVA-OneVision 72B locally at Q4, you need ~42 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).

Check Apple Mac Studio price on Amazon →

As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.

Published benchmark scores

BenchmarkScore
MMMU69.5

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

To put LLaVA-OneVision 72B in context: its MMMU score of 69.5 ranks #3 of the 8 catalog models with a published MMMU result (catalog median 62.8). Scores are author-published and measured under different harnesses, so treat gaps of a few points as noise rather than a verdict.

Strengths

  • State-of-the-art open vision quality at 2024 release
  • Robust multi-image and video reasoning
  • Apache 2.0 with no usage restrictions
  • Solid bilingual EN/CN coverage

Limitations

  • Around 42 GB VRAM at Q4 — needs serious GPU resources
  • 32k context limits long-document workflows
  • Surpassed by Qwen3-VL 30B in 2025 benchmarks

Typical workloads

In our catalog grid, LLaVA-OneVision 72B is filed under Large Open VLM, EN/CN — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: vision-language work — screenshots, charts, scanned documents.

The 32k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. The Apache 2.0 license is permissive, so shipping it inside a commercial product raises no special legal questions.

Architecture & training

Architecture: Dense vision · 72B · LLaVA-OneVision · Qwen2 72B backbone

Training: LMMs-Lab — images, videos, documents, multi-image. OneVision dataset.

Verdict

A heavyweight Apache VLM that still delivers, though Qwen3-VL has since taken the open-vision crown at lower cost.

Quick start

ollama run llava-onevision:72b

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.

Similar models worth comparing

Frequently asked questions

How much VRAM does LLaVA-OneVision 72B need?

At the recommended Q4_K_M quantization, LLaVA-OneVision 72B needs about 42 GB of VRAM. Q8_0 takes 78 GB, and unquantized FP16 weights take 144 GB.

Can LLaVA-OneVision 72B run without a GPU?

Yes — with roughly 64 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.

What context window does LLaVA-OneVision 72B support?

LLaVA-OneVision 72B supports a 32k-token context window (32,768 tokens).

Can I use LLaVA-OneVision 72B commercially?

Yes. LLaVA-OneVision 72B is released under Apache 2.0, a permissive open-source license that allows commercial use, modification and redistribution.

How fast is LLaVA-OneVision 72B on consumer hardware?

Our compatibility engine estimates on the order of 5 tokens/sec on a mid-range GPU and up to 18 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of LLaVA-OneVision 72B should I download first?

Start with Q4_K_M (42 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.

Tools

Is LLaVA-OneVision 72B the right pick for you?

Compute self-hosted ROI → Back to catalog