Llama 3.2 3B Instruct
Meta · released Sep 18, 2024 · meta-llama/Llama-3.2-3B-Instruct
- Type
- Open weightsLlama 3.2 Community License
- Params
- 3.2B
- Context
- 131K
about 98K words of context · download allowed, licence restricts use
Our take
Written Sep 2, 2026Llama 3.2 is a tiny downloadable model from Meta built for on-device and low-latency tasks. It handles up to 131,072 tokens in a single request and is extremely cheap to run, though its knowledge and reasoning scores lag well behind larger alternatives.
Pick this for edge deployment where memory is tight, or for long-context tasks on a budget at the smallest size class. Use it when instruction-following matters more than deep knowledge. Skip it if you need graduate-level science reasoning, broad factual accuracy, or guaranteed fast throughput — speed is unmeasured on the cheapest host.
The case for it
- Cheapest listed rate is several times lower than the most expensive host for the same weights.
- 131,072-token request limit is unusually long for a 3.2-billion-parameter model.
- 72% instruction-following accuracy on a formal rubric.
The case against it
- Near-random on graduate science questions: under 3% correct.
- Broad knowledge is weak at one-fifth correct on a standard academic test.
- Output price spans more than sixfold across providers for identical weights; throughput is undisclosed on the cheapest.
How good is it?
A small open text model for basic chat and simple prompts, though it trails most others at everyday questions, writing and code.
- answering everyday questionsArena Text (overall) · 167th of 168
- drafting and editing textArena Creative Writing · 167th of 168
- writing and completing codeArena Coding · 167th of 168
EverydayGeneral questions and everyday reasoning
Arena Text (overall)167th of 168 · 1167
CodingWriting and fixing code on its own
Arena Coding167th of 168 · 1177
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent.
WritingDrafting and rewriting prose
Arena Creative Writing167th of 168 · 1145
Arena Creative Writing is the only board that has scored it for this.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 19.5 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 27.5 GB spare means a 10% error in the size would not change the answer.
Apple M2 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 2.7 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 4 hours and 21 days ago — each listing carries its own date.
The only listing at 131K of context — the other 3 in the table below are not like-for-like. One cheaper row there is outside that comparison: a different context length.
- per 1M tokens
- $0.050 in / $0.33 out
- Context served
- 131K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Novita AIDirect | $0.030 / $0.050checked 21 days ago | 33K | not measured | Unknown | Unknown | Unknown |
| OpenRouterOpenRouter's own listing | $0.050 / $0.33checked 4 hours ago | 131K | not measured | Unknown | Unknown | Unknown |
| Parasailbf16Through OpenRouter | $0.050 / $0.33checked 4 hours ago | 131K118K max reply | 75 tok/s | No | No | Confirmed |
| Cloudflare Workers AIThrough OpenRouter | $0.051 / $0.34checked 4 hours ago | 80K72K max reply | 94 tok/s | No | Yesunknown period | Unknown |
Novita AI
Cloudflare Workers AIAcross the 4 listings we hold: 2 say they do not train on prompts, 0 say they do and 2 do not say. 1 appears in the zero-retention registry we check; the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| Novita AIDirect | |||
| OpenRouterOpenRouter's own listing | ✗ | ✗ | ✓ |
| Parasailbf16Through OpenRouter | ✗ | ✗ | ✓ |
| Cloudflare Workers AIThrough OpenRouter | ✗ | ✗ | ✗ |
Tool calling: 0 of 4 listings say yes, 3 say no, 1 publishes no parameter list. JSON output: 0 of 4 listings say yes, 3 say no, 1 publishes no parameter list. Strict schema: 2 of 4 listings say yes, 1 says no, 1 publishes no parameter list.
Models people weigh against Llama 3.2 3B Instruct
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- 1 of 4 listings publishes no parameter list, so what its API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 4 listings do not say whether they train on prompts.
- We hold no cached-input rate for any of its listings.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsLlama 3.2 Community License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Llama 3.2 Community License
Same terms as Llama 3.1 (700M MAU cap, naming rules). The multimodal 3.2 models add a clause restricting use by entities domiciled in the EU.
Identifiers
- Hugging Face
- meta-llama/Llama-3.2-3B-Instruct
- Takes in, gives back
- Text in, text out
- Catalogue slug
- meta-llama-llama-3-2-3b-instruct