Cohere has launched North Mini Code, a new open weights coding model with 30 billion total parameters and 3 billion active parameters, achieving a score of 27.6 on the Artificial Analysis Intelligence Index. This model, released less than a month after the Command A+ model, is designed to optimize coding tasks and outperforms other models in its size class, such as GLM-4.7-Flash, on the Artificial Analysis Coding Index with a score of 33.4. However, it shows lower performance in non-coding tasks. Open-sourced under the Apache 2.0 license, North Mini Code runs faster than many comparable models on Cohere’s API, demonstrating around 199 output tokens per second.

Cohere: Cohere is an AI company that develops large language models tailored for enterprise and developer use cases. It recently released North Mini Code, a compact open-weights coding model, shortly after launching Command A+.
North Mini Code: North Mini Code is Cohere’s newly released open-weights model optimized for coding tasks. It is a text-only model released under the Apache 2.0 license and designed to deliver strong performance in its size class on coding benchmarks.

API Speed: The model runs faster on Cohere’s API than several comparable open weights models of similar intelligence and size.
Licensing: North Mini Code is released under the Apache 2.0 license to support broad accessibility and community use.
Benchmark Performance: North Mini Code scores competitively on coding evaluations relative to other open weights models in its size class while showing relative weakness on certain non-coding agentic tasks.