North Mini Code 1.0
30.5B parameter open-weight model. Cohere Labs' first agentic coding model and the debut of its North family. Sparse mixture-of-experts: 30.5B total parameters across 128 experts, 8 active per token. The Ollama library tag ships Q4_K_M at ~18 GB. Measured on an RTX 3080 (10 GB) it ran split across GPU and CPU (roughly 46/54 by resident bytes, ~8.9 GB in VRAM) and still sustained ~30 tok/s — usable on a 10 GB card, comfortable on 24 GB. The 500K context window is the largest in this catalogue. Apache 2.0 licensed, so unencumbered for commercial use.
Cohere Labs · North
Model overview
North Mini Code is Cohere Labs' first agentic coding model. Like Laguna XS it is a mixture-of-experts design — 30.5B parameters total, 8 of 128 experts active per token — which keeps inference cost near that of a much smaller model. It supports tool calling and an explicit reasoning mode, and its 500K context window is large enough to hold an entire repository plus history. In hands-on testing it produced correct, well-structured code for a standard interval-merging problem with working test cases, and stayed on topic across a long response. Apache 2.0 licensing makes it one of the least encumbered options at this scale.
Editorial review
VRAM figures are empirical estimates. Actual usage varies by runtime, context length, and system configuration. Verify on your specific hardware before production use.
Will North Mini Code 1.0 run on your machine?
North Mini Code 1.0 is 30.5B parameters and needs 19.5 GB of VRAM at Q4_K_M — 18 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.
VRAM by quantization
| Quantization | Weights | Needs (with overhead) | Quality |
|---|---|---|---|
| Q4_K_M | 18 GB | 19.5 GB | good |
Fit on common hardware at Q4_K_M
| Hardware | Memory the model can use | System RAM | Verdict |
|---|---|---|---|
| CPU Only | None (CPU only) | 16 GB | Too large |
| RTX 4060 Laptop | 8 GB | 16 GB | Too large |
| RTX 3060 (12GB) | 12 GB | 32 GB | CPU offload |
| RTX 4060 Ti (16GB) | 16 GB | 32 GB | CPU offload |
| RTX 3090 | 24 GB | 64 GB | Comfortable |
| Apple Silicon (Unified Memory) 36 GB | 27 GB of 36 GB | 36 GB | Comfortable |
| RTX 5090 | 32 GB | 64 GB | Comfortable |
Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.
Apple Silicon shares one pool of memory between the system and the GPU, so a model cannot use all of it. These rows apply the same 75% usable fraction the Compatibility Checker uses, which is why a 36 GB Mac is graded on less than 36 GB.
Ready to run this model locally?
Find a compatible interface in our Local AI Tools directory →