Alibaba (Qwen) · Apache 2.0
Qwen3-Coder
Qwen models made for coding agents. The 30B-A3B fits a 24 GB card and answers fast.
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 |
|---|---|---|---|---|---|
| 31B (3.3B active) | Q4_K_M | 18.6 GB | 20.4 GB | 22.8 GB | 32.4 GB |
| Q8_0 | 32.5 GB | 34.3 GB | 36.7 GB | 46.4 GB | |
| 80B (3B active) | Q4_K_M | 48.5 GB | 50.3 GB | 52.8 GB | 62.4 GB |
| Q8_0 | 84.8 GB | 86.6 GB | 89.0 GB | 98.7 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 | 31B Q4_K_M | Just fits | 12–25tok/sSmooth |
| RTX 4060 · 8 GB | 31B Q4_K_M | Spills into system RAM | 14–30tok/sSmooth |
| RTX 3060 · 12 GB | 31B Q4_K_M | Spills into system RAM | 19–35tok/sSmooth |
| RTX 5060 Ti · 16 GB | 31B Q4_K_M | Spills into system RAM | 30–55tok/sFast |
| RTX 4090 · 24 GB | 31B Q4_K_M | Runs well | 90–180tok/sFast |
| RTX 5090 · 32 GB | 31B Q4_K_M | Runs great | 130–250tok/sFast |
| Mac with M4 · 16 GB | 31B Q4_K_M | Too big | — |
| Mac with M4 Pro · 48 GB | 31B Q8_0 | Runs well | 25–55tok/sFast |
| Mac with M5 Max · 128 GB | 80B Q8_0 | Runs great | 45–100tok/sFast |
| Ryzen AI Max+ 395 · 128 GB | 80B Q8_0 | Runs well | 25–45tok/sSmooth |
| DGX Spark · 128 GB | 80B Q8_0 | Runs well | 25–45tok/sSmooth |
| RTX PRO 6000 · 96 GB | 80B Q8_0 | Runs well | 100–200tok/sFast |
| Mac Studio with M5 Ultra · 256 GB | 80B Q8_0 | Runs great | 45–95tok/sFast |
| Mac Studio with M3 Ultra · 512 GB | 80B Q8_0 | Runs great | 35–75tok/sFast |
Measured speeds
Real runs, not estimates. All measurements.
| Computer | Version | Prompt | Speed | Source |
|---|---|---|---|---|
| NVIDIA DGX Spark (GB10, 128 GB LPDDR5x) | Q8_0 | 512 | 61.06 tok/s | 2026-02-05 |
What it is good and bad at
Good at
- Coding agents and tool calls
- Fast: only about 3B parameters work per word
Not good at
- 30B-A3B conversation memory grows quickly with long context
- Not a general chat model
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 qwen3-coder:30bollama run qwen3-coder-next:latest