Meta · Llama 3.1–3.3 Community
Llama 3
Meta’s 2024 models. Older and weaker than 2026 models, but supported by every tool.
How much memory it needs
The model file, plus memory for the conversation, plus about 1 GB for the app. Longer conversations need more.
| Size | Version | File | At 8K | At 32K | At 128K |
|---|---|---|---|---|---|
| 1.2B | MXFP4 | 0.8 GB | 2.1 GB | 2.9 GB | 6.1 GB |
| 3.2B | Q4_K_M | 2.0 GB | 4.0 GB | 6.8 GB | 18.1 GB |
| Q8_0 | 3.4 GB | 5.4 GB | 8.2 GB | 19.5 GB | |
| 8B | Q4_K_M | 4.9 GB | 7.0 GB | 10.2 GB | 23.1 GB |
| Q8_0 | 8.5 GB | 10.6 GB | 13.8 GB | 26.7 GB | |
| 71B | Q4_K_M | 42.5 GB | 46.2 GB | 54.3 GB | 86.5 GB |
| Q8_0 | 75.0 GB | 78.7 GB | 86.7 GB | 119 GB |
Which computers run it
The best size for each machine at an 8K-token conversation. Speeds are estimates from memory bandwidth.
| Computer | Best version | Fit | Speed |
|---|---|---|---|
| Laptop or PC, no graphics card, 32 GB | 8B Q8_0 | Runs great | 5.5–9tok/sReadable |
| RTX 4060 · 8 GB | 8B Q4_K_M | Runs well | 25–40tok/sSmooth |
| RTX 3060 · 12 GB | 8B Q8_0 | Runs well | 20–35tok/sSmooth |
| RTX 5060 Ti · 16 GB | 8B Q8_0 | Runs great | 30–45tok/sSmooth |
| RTX 4090 · 24 GB | 8B Q8_0 | Runs great | 60–95tok/sFast |
| RTX 5090 · 32 GB | 8B Q8_0 | Runs great | 100–160tok/sFast |
| Mac with M4 · 16 GB | 8B Q4_K_M | Runs great | 12–20tok/sSmooth |
| Mac with M4 Pro · 48 GB | 8B Q8_0 | Runs great | 16–30tok/sSmooth |
| Mac with M5 Max · 128 GB | 71B Q4_K_M | Runs great | 8–13tok/sReadable |
| Ryzen AI Max+ 395 · 128 GB | 8B Q8_0 | Runs great | 16–25tok/sSmooth |
| DGX Spark · 128 GB | 8B Q8_0 | Runs great | 17–25tok/sSmooth |
| RTX PRO 6000 · 96 GB | 71B Q8_0 | Runs well | 14–20tok/sSmooth |
| Mac Studio with M5 Ultra · 256 GB | 71B Q8_0 | Runs great | 7–12tok/sReadable |
| Mac Studio with M3 Ultra · 512 GB | 71B Q8_0 | Runs great | 5–8tok/sReadable |
Measured speeds
Real runs, not estimates. All measurements.
| Computer | Version | Prompt | Speed | Source |
|---|---|---|---|---|
| Apple M4 Pro (20-core GPU, 24 GB) | MLX 4-bit | 4K | 47.6 tok/s | 2026-10-06 |
| Apple M4 Max (12P+4E CPU, 40-core GPU, 128 GB) | Q4_K_M | 2K | 52.9 tok/s | |
| NVIDIA GeForce RTX 5060 Ti (reported as 15 GB) | Q4_K_M | 2K | 59 tok/s |
What it is good and bad at
Good at
- Works everywhere
- Many fine-tuned versions exist
Not good at
- Clearly behind 2026 models of the same size
- 70B needs 48 GB or more
- Licence has extra terms for very large companies
Start it
Install Ollama, then run one of these in a terminal. LM Studio users can search the same name in the app.
ollama run llama3.2:3bollama run llama3.1:8bollama run llama3.3:70b