NEWS · MODELS · #398
Cohere releases North Mini Code — open-source 30B (3B active) agentic coding MoE model
Cohere has open-sourced North Mini Code, a mixture-of-experts agentic coding model sized at 30 billion parameters with 3 billion active, released under an Apache 2.0 license. The model — available on Hugging Face, OpenRouter, and Cohere's Model Vault in bf16, fp8 and w4a16 formats and compatible with OpenCode — is optimized for agentic developer workflows and reports a 33.4 score on the Artificial Analysis Coding Index, higher throughput (up to 2.8x) and ~30% better inter-token latency versus Devstral Small 2 in Cohere's tests.
KEY POINTS
- Cohere has open-sourced North Mini Code, a mixture-of-experts agentic coding model sized at 30 billion parameters with 3 billion active, released under an Apache 2.0 license.
- The model — available on Hugging Face, OpenRouter, and Cohere's Model Vault in bf16, fp8 and w4a16 formats and compatible with OpenCode — is optimized for agentic developer workflows and reports a 33.4 score on the Artificial Analysis Coding Index, higher throughput (up to 2.8x) and ~30% better inter-token latency versus Devstral Small 2 in Cohere's tests.
- An open-source, efficient agentic coding MoE from a major model provider lowers barriers for on‑prem and sovereign developer tooling and could shift adoption toward smaller, deployable coding agents.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Cohere published North Mini Code, an open-source mixture-of-experts (MoE) agentic coding model: 30B total parameters with 3B active, released under an Apache 2.0 license and distributed on Hugging Face, OpenRouter, and Cohere’s Model Vault in bf16, fp8 and w4a16 formats (compatible with OpenCode). Cohere reports benchmark and performance measurements versus Devstral Small 2 (33.4 Artificial Analysis Coding Index; up to 2.8× throughput; ~30% better inter-token latency).
WHY NOW
Cohere frames North Mini Code as part of a broader push to enable ‘sovereign’ and agentic developer tooling: it is their first open-source agentic coding model and is released alongside other North family offerings (e.g., North Small Translate and North Automations), signaling an effort to supply models and orchestration for developer and enterprise agentic workflows.
WHO IS AFFECTED
Directly affected: developers and the ‘sovereign developer’ ecosystem (explicitly named by Cohere). Also relevant to enterprises building or orchestrating agentic workflows (Cohere’s North Automations targets enterprise agent orchestration).
CONFIRMED
All items are directly stated in Cohere’s excerpts: - North Mini Code is open-source, a mixture-of-experts (MoE) agentic coding model. - Model size: 30B total parameters with 3B active. - License: Apache 2.0. - Distribution: weights available on Hugging Face; deployable on Cohere Model Vault; usable via OpenRouter and with a Cohere API key. - Formats: bf16, fp8, w4a16. - Compatibility: compatible with OpenCode. - Cohere reports a 33.4 score on the Artificial Analysis Coding Index for North Mini Code. - Cohere reports up to 2.8× higher output throughput and ~30% better inter-token latency versus Devstral Small 2 in their tests, with Devstral Small 2 showing a slight edge on time-to-first-token. - Cohere describes North Mini Code as its first open-source agentic coding model and emphasizes efficiency and lower hardware demand. (Sources: 460; contextual product-family releases and orchestration described in 471 and 498.)
UNCERTAIN
Missing or unresolved in the supplied excerpts: - Independent third-party verification of Cohere’s benchmark and performance claims (throughput, latency, Artificial Analysis Coding Index) is not provided in the excerpts — results are Cohere’s internal tests. - Broad comparative position vs. other coding models beyond the single comparison to Devstral Small 2 is not shown. - Operational details for running the MoE (e.g., hardware configurations required to obtain the reported 3B active behavior, tooling/complexity for on‑prem MoE inference) are not specified. - Security, safety, and governance implications of deploying an open-source agentic coding model are not detailed. - No conflicting evidence is present in the supplied excerpts; the absence of independent or external data is the main evidence gap.
WHAT TO WATCH
Concrete observable signals to monitor next (all items are traceable to channels named by Cohere or standard public metrics): - Independent benchmark publications or replication of the Artificial Analysis Coding Index and throughput/latency comparisons vs Devstral Small 2 or other coding models. - Hugging Face: download counts, model forks, issues/pull requests, and community demos for North Mini Code. - OpenRouter / Model Vault / OpenCode: number of deployments, example integrations, or published deployment guides. - Cohere communications: updates, implementation guides, or notes on recommended hardware and inference recipes. - Community engagement: posts, tags, or demonstrations on X, Discord, Reddit as referenced by Cohere. - Any third‑party security/audit reports or enterprise case studies describing on‑prem or sovereign deployment experiences.
WHY IT MATTERS
An open-source, efficient agentic coding MoE from a major model provider lowers barriers for on‑prem and sovereign developer tooling and could shift adoption toward smaller, deployable coding agents.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
Cohere has open-sourced North Mini Code under an Apache 2.0 license.
SUPPORTEDChecked 2026-09-25 · 1 supporting
North Mini Code is a mixture-of-experts (MoE) coding model with 30 billion total parameters and 3 billion active.
SUPPORTEDChecked 2026-09-25 · 1 supporting
Cohere provides model weights on Hugging Face and Model Vault and says the model can be tried via OpenCode, OpenRouter, or a Cohere API key; weights are offered in bf16, fp8, and w4a16 formats.
SUPPORTEDChecked 2026-09-25 · 1 supporting
Cohere reports North Mini Code achieves a 33.4 score on the Artificial Analysis Coding Index.
SUPPORTEDChecked 2026-09-25 · 1 supporting
SOURCES & TIMELINE
1Small, efficient, and open-source — our first agentic coding model, built for the sovereign developer ecosystem. Today we're launching North Mini Code open-source. A mixture-of-experts (MoE) model, North Mini Code is Cohere's first agentic coding model, and the inaugural member of our next generation of powerful models. At 30B total parameters with just 3B active, North Mini Code delivers strong software development …
Outsized performance, right-sized footprint — translation built for speed and cost efficiency. Today, we're releasing North Small Translate, a mixture-of-experts machine translation model with strong performance across 50+ languages. Across WMT26 benchmarks,¹ North Small Translate achieves an 83.6 score across all languages, outperforming proprietary models like DeepL and Google Translate, as well as open-weight alte…
We're excited to announce the launch of North Automations , Cohere's latest step in enabling enterprises to close the AI ROI gap with ease, control, and governance built in from the ground up. The rapid proliferation of AI agents across the enterprise has created a new set of challenges that are widening the gap between agentic AI potential and its practical implementation. These challenges include: Narrow workstream…