Moonshot AI · MIT
Kimi Linear 48B-A3B
Moonshot’s long-context model: 1M tokens with a tiny conversation memory.
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 |
|---|---|---|---|---|---|
| 49B (3B active) | Q4_K_M | 30.1 GB | 31.1 GB | 31.3 GB | 32.1 GB |
| Q8_0 | 52.3 GB | 53.3 GB | 53.5 GB | 54.3 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 | 49B Q4_K_M | Too big | — |
| RTX 4060 · 8 GB | 49B Q4_K_M | Too big | — |
| RTX 3060 · 12 GB | 49B Q4_K_M | Spills into system RAM | 20–40tok/sSmooth |
| RTX 5060 Ti · 16 GB | 49B Q4_K_M | Spills into system RAM | 25–50tok/sSmooth |
| RTX 4090 · 24 GB | 49B Q4_K_M | Spills into system RAM | 45–90tok/sFast |
| RTX 5090 · 32 GB | 49B Q4_K_M | Just fits | 150–300tok/sFast |
| Mac with M4 · 16 GB | 49B Q4_K_M | Too big | — |
| Mac with M4 Pro · 48 GB | 49B Q4_K_M | Runs well | 45–95tok/sFast |
| Mac with M5 Max · 128 GB | 49B Q8_0 | Runs great | 50–110tok/sFast |
| Ryzen AI Max+ 395 · 128 GB | 49B Q8_0 | Runs great | 25–55tok/sFast |
| DGX Spark · 128 GB | 49B Q8_0 | Runs great | 30–55tok/sFast |
| RTX PRO 6000 · 96 GB | 49B Q8_0 | Runs great | 120–230tok/sFast |
| Mac Studio with M5 Ultra · 256 GB | 49B Q8_0 | Runs great | 50–110tok/sFast |
| Mac Studio with M3 Ultra · 512 GB | 49B Q8_0 | Runs great | 40–90tok/sFast |
What it is good and bad at
Good at
- Very long documents
- Fast: 3B parameters work per word
Not good at
- No Ollama library entry
- Only one trusted GGUF builder