Skip to content
AI Model Radar
IndexCollected 1 h ago

Model · Meta

Llama 3.2 3B

3B · Q4_K_M · 1.9 GB · open weights · curated for general · repo created 2024-09-18

EstimatedEstimated: computed from our curated model and hardware catalog — not a live reading.Hugging Face repo (External link)GGUF file (External link)Ollama (External link)

On the radar

76

Heat Score · rising ▲ · 100% confidence

Rank among scored models
5 of 45
Trending score · Hugging Face
24
Downloads, rolling 30 days · Hugging Face
1,605,242
Downloads vs. the reading of 2026-09-05
+11.8%
Likes · Hugging Face
2,563
API price per 1M tokens, in / out · OpenRouter
$0.049999999999999996 / $0.33

Measured signals as of 2026-09-12 12:00 UTC · How the Heat Score is made →

Downloads, daily readingsdaily last · UTC

2026-08-31 · 1.1M2026-09-12 · 1.6M

Heat is a composite of measured signals only: each component is the model's percentile among models with a full week of history — trending level (35%), 7-day download growth (30%), 7-day trending change (15%), 30-day downloads (15%) and Hub likes (5%). Missing components renormalize the weights and lower the shown confidence; nothing is guessed. Trending and downloads come from the Hugging Face Hub — the download counter is a rolling 30-day window, not unique users — and input pricing from OpenRouter (CC BY 4.0). The full formula, thresholds and flag rules are on the methodology page.

What it needs

EstimatedEstimated: computed from our curated model and hardware catalog — not a live reading.

Weights, context cache and runtime overhead at four reference contexts. The total is the model's own; what differs per machine is the usable memory it has to fit into.

The attention shape of this model is unpublished, so the context cache cannot be estimated. Verdicts stay UNKNOWN rather than guessed.

All memory figures are estimates: measured quantized file size + computed context memory + runtime overhead, with a 12% safety margin on your hardware. Real usage varies with runtime version and settings.

Where it runs

EstimatedEstimated: computed from our curated model and hardware catalog — not a live reading.

Every tracked GPU and Mac at 8K and 32K context with a full-precision (f16) cache — the same engine as the finder and the GPU pages.

The attention shape of this model is unpublished, so the context cache cannot be estimated. Verdicts stay UNKNOWN rather than guessed.

Where it runs
Device8K context32K context
RTX 407012 GB · 10.6 GB usableUnknownUnknown
RTX 3060 12GB12 GB · 10.6 GB usableUnknownUnknown
RTX 408016 GB · 14.1 GB usableUnknownUnknown
RTX 508016 GB · 14.1 GB usableUnknownUnknown
RTX 4060 Ti 16GB16 GB · 14.1 GB usableUnknownUnknown
RTX 5060 Ti 16GB16 GB · 14.1 GB usableUnknownUnknown
RTX 5070 Ti16 GB · 14.1 GB usableUnknownUnknown
RTX 309024 GB · 21.1 GB usableUnknownUnknown
RTX 409024 GB · 21.1 GB usableUnknownUnknown
RTX 509032 GB · 28.2 GB usableUnknownUnknown
Mac mini M4 Pro 64GB64 GB unified · 44.8 GB usableUnknownUnknown
Mac Studio M4 Max 64GB64 GB unified · 44.8 GB usableUnknownUnknown
Mac Studio M3 Ultra 96GB96 GB unified · 67.2 GB usableUnknownUnknown
NVIDIA DGX Spark128 GB unified · 112.6 GB usableUnknownUnknown

GGUF is the quantized single-file format local runtimes load (Ollama, LM Studio, llama.cpp); our memory figures are measured from the linked GGUF file. The original repo holds full-precision safetensors — a much larger download meant for GPUs with far more memory.

Check it against your own machine →

Or rent it

Rental fit cannot be estimated without the model's attention shape.

Live rental prices →Buy or rent? The guide →

More from Meta

Every tracked model →