whatmodel.sh / Llama 3.2 90B Vision Instruct

Llama 3.2 90B Vision Instruct

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Llama 3.2 90B Vision Instruct is an open-weights language model from Meta, released September 2024. Its published size is 88.59B parameters and a 128k-token context window. Licence: llama3.2.

Catalogue identity notes

Comparison model from the Epoch AI internal-evaluation CSVs (Epoch release date 2024-09-24). Vision-language instruct checkpoint Llama-3.2-90B-Vision-Instruct (text benchmarks here). 88,593,355,323 parameters (safetensors metadata, includes the vision adapter).

49.9±1.3#342 of 719 LLMsfrom 13 benchmark results across 7 sources

$ how it compares

The LLMs ranked either side of Llama 3.2 90B Vision Instruct, out of 828 with a score, then more from Meta.

#modelscorewhereRAMreleased
410Mellum2 12B-A2.5B Instruct JetBrains50±2local~8.6GB2026-05
411OpenCodeInterpreter-DS-33B M-A-P50±1local~22GB2024-02
412Llama 3.2 90B Vision Instruct Meta50±1local~58GB2024-09
413DeepSeek-Coder 33B Instruct DeepSeek50±1local~21GB2023-11
414Yi 34B 200K (base) 01.AI50±5local~23GB2023-11
more from Meta
24Muse Spark 1.3 Meta77±1api $2.00/M2026-09
40Muse Spark 1.2 Meta75±1api $2.00/M2026-08
43Muse Spark 1.1 Meta75±1api $2.00/M2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

math▁▂▃▅▆▅▄█▄▁▁▁ #274 of 398 · 40% of its score · 3 tests

Beats Molmo 2 8B, DeepSeek-Coder-V2 Instruct (0724), Llama 3.1 70B Instruct here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Granite 8B Code Base (4K), Mixtral 8x7B v0.1 (base) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
MATH level 539.4% ✎▂▂▅▆▄█▆▃▅▂▃▄ beats 31/92hard · avg 32%▇▇▇▇▇▇▇▇▇▇▇▇▇ 32%sharp · at its level
OTIS mock AIME 2024–20252.6% ✎▃▃▄▃▃▃▇▇█▃▂▅ beats 22/193very hard · avg 8.5%▇▇ 4.6%sharp · noisy
MathVista (testmini)49.1% ✎▂····▂···▂█▆ beats 3/9medium · avg 53%▇ 3.4%at its level
vision & documents▁▁▃▃▁▃▄▂▆▇█▄ #125 of 147 · 32% of its score · 7 tests

Trails Qwen3.5 2B, Pixtral 12B (2409), Qwen3-VL 2B Thinking here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena vision (style controlled)1032.3 ✎▁▁▂▂▃▂▄▅▅▆██ beats 19/145avg 1068▇▇▇▇ 10%blunt
CapArena-Auto detailed image captioning (Score_Avg, GPT-4o judge)-25.6700 ✎▃▃▃·▃█·▅▃▃▃▃ beats 3/12avg -0.671▇▇▇ 6.9%typical
MM-MT-Bench (Mistral, GPT-4o judge)5.5000 MM-MT-Bench score (1-10) ✎▅··▃·█·▃▃··▅ beats 5/9avg 5.740 MM-MT-Bench score (1-10)▇▇ 5.0%sharp · noisy
MMMU (val)53.7% ✎▂·▂······█▄▄ beats 6/9medium · avg 52%▇ 3.6%at its level
MathVista (testmini)49.1% ✎▂····▂···▂█▆ beats 3/9medium · avg 53%▇ 3.4%at its level
VQAv2 (val)67.0% ✎▃···█▃··▅▃▃▃ beats 1/9easy · avg 74%▇ 2.9%blunt · at its level
ChartQA33.8% ✎▄·▄▄▄··▄▄▄▄█ beats 2/9easy · avg 77%▇ 0.2%blunt · noisy
other tests4 not in a task · 30% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond41.0% ✎▁▁▁▁▂▄▆▄▅▇█▅ beats 42/208hard · avg 40%▇▇▇▇▇▇▇ 17%sharp
AraGen-v3 3C3H score (Inception, Claude Sonnet 4 judge)50.5% ✎▁▁·▁▂▂▄█▄▅▄▄ beats 44/66hard · avg 31%▇▇▇ 6.9%at its level
Open PT-LLM Leaderboard average (9 Portuguese tasks)79.3% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 253/276medium · avg 73%▇ 3.2%at its level · noisy
LongSocialBench (full-context, 4-option MCQ over discussion threads)35.6% ✎▆▂·▄█·▂▂·▄▄▂ beats 0/16very hard · avg 36%▇ 3.1%blunt

$ where to run it

Ollama · recommended

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Llama 3.2 90B Vision Instruct · needs ~53.65 GB · from ollama.com
    chat in the terminal
    ollama run llama3.2-vision:90b

Weights and documentation: huggingface.co/meta-llama ↗

$ how the score was worked out · Epoch AI — internal evaluatio… 54%, Pixtral 12B (Mistral AI, arXi… 15%, LMArena leaderboard dataset 10% and 4 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Epoch AI — internal evaluations3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 54%CC BY 4.0
Pixtral 12B (Mistral AI, arXiv 2410.07073v2)5▇▇▇▇▇ 15%CC BY 4.0
LMArena leaderboard dataset1▇▇▇ 10%CC BY 4.0
Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)1▇▇ 7%Apache-2.0
CapArena: Benchmarking and Analyzing Detailed Image Captioning in the LLM Era (arXiv:2503.12329v1)1▇▇ 7%CC-BY-4.0
Open Portuguese LLM Leaderboard (archived)1▇ 3%Apache-2.0
LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads? (arXiv:2610.04118v1)1▇ 3%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 49.9

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json