Skip to content
Local AI Wizard
Your machine
Theme

Llama 3.2 3B Instruct

Meta · meta-llama/Llama-3.2-3B-Instruct

Text & chat

Q4 file 2.0 GB (Q4_K_M) · runs well from 6 GB · Weights license: Publisher terms: Llama 3.2

LMArena · text1,167 ± 8Rank 371 · release 2026-10-22 · its own scale
Memory needed~3 GBQ4 weights + working room, before the machine's reserve
Parameters3.21 Bmanufacturer specification
Context128,000 tokensPublished for this model

Quality

LMArena · text · rating
1,167 ± 8Confidence interval 1,159–1,175 · 7,936 votes · rank 371 of the releaseLMArena 2026-10-22 · CC BY 4.0

No result in LiveBench 2026-06-25 Not in LiveBench 2026-06-25

Reported by the publisher

22 results · Meta's model card · read 2026-10-25
MMLU
63.4Llama 3.2 3B bf16 · Instruction Tuned Models · General
Open-rewrite eval
40.1Llama 3.2 3B bf16 · Instruction Tuned Models · Re-writing
TLDR9+ (test)
19.0Llama 3.2 3B bf16 · Instruction Tuned Models · Summarization
IFEval
77.4Llama 3.2 3B bf16 · Instruction Tuned Models · Instruction following
GSM8K (CoT)
77.7Llama 3.2 3B bf16 · Instruction Tuned Models · Math
MATH (CoT)
48.0Llama 3.2 3B bf16 · Instruction Tuned Models · Math
ARC-C
78.6Llama 3.2 3B bf16 · Instruction Tuned Models · Reasoning
GPQA
32.8Llama 3.2 3B bf16 · Instruction Tuned Models · Reasoning
Every reported result22
Hellaswag
69.8Llama 3.2 3B bf16 · Instruction Tuned Models · Reasoning
BFCL V2
67.0Llama 3.2 3B bf16 · Instruction Tuned Models · Tool Use
Nexus
34.3Llama 3.2 3B bf16 · Instruction Tuned Models · Tool Use
InfiniteBench/En.QA
19.8Llama 3.2 3B bf16 · Instruction Tuned Models · Long Context
InfiniteBench/En.MC
63.3Llama 3.2 3B bf16 · Instruction Tuned Models · Long Context
NIH/Multi-needle
84.7Llama 3.2 3B bf16 · Instruction Tuned Models · Long Context
MGSM (CoT)
58.2Llama 3.2 3B bf16 · Instruction Tuned Models · Multilingual
MMLU · Portuguese (5-shot, macro_avg/acc)
54.5Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · Spanish (5-shot, macro_avg/acc)
55.1Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · Italian (5-shot, macro_avg/acc)
53.8Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · German (5-shot, macro_avg/acc)
53.3Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · French (5-shot, macro_avg/acc)
54.6Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · Hindi (5-shot, macro_avg/acc)
43.3Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
MMLU · Thai (5-shot, macro_avg/acc)
44.5Llama 3.2 3B · Multilingual Benchmarks · General · 5-shot
Across Llama 3.2 sizes22 shared results · same publisher and settings
Llama 3.2: results the publisher reports for each size under the same settings
Benchmark (publisher settings)1B Instruct3B Instruct
MMLU · General49.363.4
Open-rewrite eval · Re-writing41.640.1
TLDR9+ (test) · Summarization16.819.0
IFEval · Instruction following59.577.4
GSM8K (CoT) · Math44.477.7
MATH (CoT) · Math30.648.0
ARC-C · Reasoning59.478.6
GPQA · Reasoning27.232.8
Hellaswag · Reasoning41.269.8
BFCL V2 · Tool Use25.767.0
Nexus · Tool Use13.534.3
InfiniteBench/En.QA · Long Context20.319.8
InfiniteBench/En.MC · Long Context38.063.3
NIH/Multi-needle · Long Context75.084.7
MGSM (CoT) · Multilingual24.558.2
MMLU · Portuguese (5-shot, macro_avg/acc) · General · 5-shot39.854.5
MMLU · Spanish (5-shot, macro_avg/acc) · General · 5-shot41.555.1
MMLU · Italian (5-shot, macro_avg/acc) · General · 5-shot39.853.8
MMLU · German (5-shot, macro_avg/acc) · General · 5-shot39.253.3
MMLU · French (5-shot, macro_avg/acc) · General · 5-shot40.554.6
MMLU · Hindi (5-shot, macro_avg/acc) · General · 5-shot33.543.3
MMLU · Thai (5-shot, macro_avg/acc) · General · 5-shot34.744.5

Values as printed on the card, with the publisher's own settings (card section: Benchmarks - English Text › Instruction Tuned Models; Benchmarks - English Text › Multilingual Benchmarks). Not comparable across publishers and not used in any ranking. Model card

Run it

Run it

ollama run hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
  1. Install Ollama
  2. Run the command above
  3. Ask: Explain what a context window is, in two sentences.

Runs on

Fast memory holds it 118

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 118

Within memory, outside the fast part 1

Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.

Llama 3.2 3B Instruct is a text model from Meta with 3.2B parameters. It was released on September 25, 2024 and reads up to 128,000 tokens of context. Its Q4 weights plus working room come to about 3 GB before the machine's reserve; the smallest reference machines whose fast memory holds that are Mac mini M2 · 8 GB unified (~5 GB with its reserve) and RTX 4060 · 8 GB VRAM (~5 GB with its reserve). We collected 17 downloadable GGUF and MLX files from 2 publisher sources; the most downloaded GGUF source is bartowski's. It has no result in LiveBench 2026-06-25. Its LMArena text rating is 1,167 ± 8 (release 2026-10-22), a separate human-preference scale.

Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Each card opens its hardware profile. Not a measured fit or speed.

On the reference machines

Llama 3.2 3B Instruct on the reference machines
Where Q4 sits
Mac mini M2 · 8 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB16,384 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB16,384 tokens
Mac mini M4 · 16 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~10 GB65,536 tokens~7 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~10 GB65,536 tokens
Mac mini M4 · 24 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~12 GB65,536 tokens~9 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~12 GB65,536 tokens
Mac mini M4 · 32 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens~11 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens
Mac mini M5 Pro · 64 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens~11 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens
Mac Studio M5 Max · 128 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens~11 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens
Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unifiedFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens~11 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~14 GB131,072 tokens
RTX 4060 · 8 GB VRAMFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB16,384 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB16,384 tokens
RTX 3060 · 12 GB VRAMFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB32,768 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB32,768 tokens
RTX 5080 · 16 GB VRAMFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB65,536 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB65,536 tokens
RTX 4090 · 24 GB VRAMFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB131,072 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB131,072 tokens
RTX 5090 · 32 GB VRAMFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB131,072 tokens~5 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~8 GB131,072 tokens
DGX H100 · 8 × 80 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~22 GB131,072 tokens~19 GBFast memory holds itQ8_0GGUF · 3.4 GB · needs ~22 GB131,072 tokens

The file column uses the site's quant ladder (Q8_0 to Q3_K_M, MLX 8 to 4-bit on a Mac): the largest collected file whose size × 1.2 plus the context cache and the machine's reserve fits the fast memory. Planning estimates, not measured runs.

Plan another quantization or context

Consider it for

Compare Llama 3.2 3B Instruct using its publisher card. No overall score appears in the LiveBench 2026-06-25 snapshot.

Check first

The Q4 weights plus working room come to about 3 GB from 3.21B stated parameters, before the machine's reserve (Runs on adds it per machine). It is not a measured minimum; context, batch size, runtime and quantization change actual memory use.

Versions to download

Files you can download for this checkpoint. The section opens on the most downloaded collected GGUF source, bartowski, at Q4_K_M. On a Mac it opens on the most downloaded MLX source. Sizes are the Hub's published bytes. What is Q4?

bartowski

bartowski/Llama-3.2-3B-Instruct-GGUF · Llama-3.2-3B-Instruct-Q4_K_M.gguf · 2.0 GB

Run this file

ollama run hf.co/bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
lms get https://huggingface.co/bartowski/Llama-3.2-3B-Instruct-GGUF
  • mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac

Quality scores on this page belong to the original checkpoint; these files are not scored or ranked. File sizes read 2026-10-23 · download counts read 2026-10-17 · commands: Hugging Face Hub docs: Use Ollama with any GGUF model · llama.cpp server README (-hf <user>/<model>[:quant]) · LM Studio docs: lms get with a Hugging Face URL · mlx-lm README (mlx_lm.generate --model)

Fine-tunes, adapters and other versions2,040 fine-tunes
Quantized versions
550
Fine-tunes
2,040
Adapters · merges
876 · 29

Browse them on the Hugging Face model tree · read 2026-10-23

Uncensored and abliterated versions18

Some community uploads remove the model's safety refusals. Quality may drop, the original publisher does not support them, claims about them are the uploader's, and the original license still applies. This site never recommends or ranks them.

18 derivative repositories have “uncensored” or “abliterated” in their name (read 2026-10-23).

Measured speed

No published measurement for this checkpoint

Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.

How to run

The memory needed above is the Q4 load before a machine's reserve. Formula: parameters × 0.625 bytes, plus 20% working allowance; each machine then adds its system reserve: 25% of unified or system memory, 2 to 8 GB, or 2 GB per GPU on dedicated cards. This is neither measured VRAM nor checkpoint download size.

Q4 planning estimate only. Check checkpoint format, runtime support, KV cache, context, batch size and offload before choosing hardware.

Evidence level: Generic estimate only; no execution reproduced by this site.

Publisher sampling defaults
temperature 0.6 · top_p 0.9generation_config.json · read 2026-10-23source
Chat template markers
tool calls · no thinking markersmarkers found in the template, not a test of the feature · read 2026-10-23source
Before downloading
  1. Check the exact checkpoint and license linked here.
  2. Confirm a compatible runtime and precision or quantization for your platform.
  3. Allow for context, batch size and any extra media components beyond the weights.

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Task and modality
language and reasoning · text
Count and basis
3.21B · manufacturer specificationmeta-llama/llama-models Llama 3.2 MODEL_CARD.md (GitHub, publisher's own card; the Hub repo is gated) · read 2026-10-17source
Context
128,000 tokens published for this modelmeta-llama/llama-models Llama 3.2 MODEL_CARD.md (GitHub, publisher's own card; the Hub repo is gated) · read 2026-10-17source
Architecture
Densemeta-llama/llama-models Llama 3.2 MODEL_CARD.md (GitHub, publisher's own card; the Hub repo is gated) · read 2026-10-17source
Released
Introducing Llama 3.2 (Meta AI blog), September 25, 2024: 1B and 3B models available for download · read 2026-10-10source
Runtime and precision
No local execution verified.
Languages
en, de, fr, it, pt, hi, es, thModel card metadata · read 2026-10-25source
Commercial use
Allowed with conditions Over 700M monthly users needs a Meta license; show "Built with Llama"; multimodal versions are not licensed to people or companies based in the EU (Acceptable Use Policy) · Llama 3.2 Community License (meta-llama/llama-models) · read 2026-10-18license text
Every stored field12 fields
Revision checked
0cb88a4f764b7a12671c53f0838cd831a0843b95
Architecture
LlamaForCausalLM
Library
transformers
Hub task tag
text-generation
Formats
safetensors, pt
Access
Gated (manual)
Publisher kind
Official
Tensor evidence
floating-tensor-elements · 3.21B · BF16checked 2026-09-26
Repository created
2024-09-18creation is not the model's release
Repository updated
2024-10-24a commit is not a release date
Hub downloads (30 days) · likes
1,106,217 · 2,875this repository only · read 2026-10-25
Catalog state
Hub config or weights found; no inventory benchmark link

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Catalog reviewed: 2026-10-17. Benchmark release: 2026-06-25; snapshot collected: 2026-09-26.