DeepReinforce (ornith-ai) · MIT
Ornith 1.5
Coding-agent models built on Qwen3.5 and Gemma 4, in a 9B and a fast 35B-A3B.
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 |
|---|---|---|---|---|---|
| 9.7B | Q4_K_M | 5.8 GB | 7.0 GB | 7.9 GB | 11.1 GB |
| Q8_0 | 9.8 GB | 11.1 GB | 11.9 GB | 15.1 GB | |
| 36B (3B active) | Q4_K_M | 21.7 GB | 22.9 GB | 23.4 GB | 25.4 GB |
| Q8_0 | 37.8 GB | 39.0 GB | 39.5 GB | 41.5 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 | 9.7B Q8_0 | Runs great | 5.5–8.5tok/sReadable |
| RTX 4060 · 8 GB | 9.7B Q4_K_M | Runs well | 25–40tok/sSmooth |
| RTX 3060 · 12 GB | 9.7B Q4_K_M | Runs great | 35–55tok/sFast |
| RTX 5060 Ti · 16 GB | 9.7B Q8_0 | Runs great | 25–40tok/sSmooth |
| RTX 4090 · 24 GB | 9.7B Q8_0 | Runs great | 55–90tok/sFast |
| RTX 5090 · 32 GB | 36B Q4_K_M | Runs great | 150–290tok/sFast |
| Mac with M4 · 16 GB | 9.7B Q4_K_M | Runs great | 11–19tok/sSmooth |
| Mac with M4 Pro · 48 GB | 36B Q4_K_M | Runs great | 45–95tok/sFast |
| Mac with M5 Max · 128 GB | 36B Q8_0 | Runs great | 50–110tok/sFast |
| Ryzen AI Max+ 395 · 128 GB | 36B Q8_0 | Runs great | 25–55tok/sFast |
| DGX Spark · 128 GB | 36B Q8_0 | Runs great | 30–55tok/sFast |
| RTX PRO 6000 · 96 GB | 36B Q8_0 | Runs great | 120–230tok/sFast |
| Mac Studio with M5 Ultra · 256 GB | 36B Q8_0 | Runs great | 50–110tok/sFast |
| Mac Studio with M3 Ultra · 512 GB | 36B Q8_0 | Runs great | 40–85tok/sFast |
What it is good and bad at
Good at
- Coding agents
- Very popular: millions of downloads a month
Not good at
- Ollama still serves the older 1.0
- Its licence file link is broken (card says MIT)