North Mini Code by Cohere is a Mixture-of-Experts model trained for agentic software engineering, terminal tasks, and code generation. Apache 2.0 licensed.
Capabilities, design details, and architectural traits
North Mini Code is a Mixture-of-Experts model designed specifically for agentic software engineering. It is the inaugural member of Cohere's North family of code agent models, released under the Apache 2.0 license for open-source, sovereign deployment.
| Trait | Detail |
|---|---|
| Sparse MoE architecture | 128 experts with 8 activated per token; sigmoid router activation before top-k selection; single dense layer before sparse layers |
| Interleaved attention | Sliding-window attention with RoPE and global attention with no positional embeddings, in a 3:1 ratio |
| Multi-harness training | Trained against multiple agent scaffolds rather than tuned to a single one, so performance generalizes across harnesses like SWE-Agent and OpenCode |
| Cascaded post-training | Two-stage supervised fine-tuning followed by reinforcement learning with verifiable rewards (RLVR), targeting software engineering and terminal tasks |
| Agentic coding focus | Built for repo-level code changes, terminal-based agents driving shell tools end-to-end, and orchestrating sub-agents across multi-turn tasks |
Cohere frames North Mini Code as a step toward sovereign AI for developers who need control over their agentic coding infrastructure. Weights are freely available in bf16, fp8, and w4a16 formats.
In internal testing, North Mini Code achieved up to 2.8x higher output throughput than Devstral Small 2 under identical concurrency, along with a 30% advantage in inter-token latency. It scores 33.4 on the Artificial Analysis Coding Index, outperforming several larger open-source models in its size class.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Cohere | Anthropic | Anthropic |
| Release Date | June 9, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 256K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $5 | $10 |
| Output Pricing | Free Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | textimage | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 20.2 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | 36.5 | 78.0 Best Coding Index | 76.5 |
| Agentic Index | 3.1 | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
