Cohere: North Mini Code (free)

Cohere: North Mini Code (free) is a text model from Cohere, listed on api14 under the model id cohere/north-mini-code:free. It is described as Cohere's first agentic coding model and the debut of the North family. Architecturally it is a sparse mixture-of-experts model with 30B total parameters and 3B active, which means only a fraction of the weights are engaged per token. The provider release date is 2026-06-17.

The model accepts text input and returns text output, so it fits text-only pipelines rather than multimodal ones. Its context window is 256,000 tokens, and the maximum output is 64,000 tokens. That combination of a large context and a large output budget matters for coding work, where a request may carry a long file, a repository excerpt or a lengthy instruction set, and the response may need to be a complete file or a multi-step plan rather than a short answer. The supported parameters include reasoning and include_reasoning, so the model can expose a reasoning mode alongside its normal output.

Because it is positioned as an agentic coding model, the natural use cases are code generation, code explanation, refactoring and debugging, and multi-step tasks where the model decides what to do next. The tool calling parameters (tools and tool_choice) make it suitable for agent loops that read files, run commands or query external services, with the model choosing when and how to call a tool. The large context window supports working across several files or a long conversation history without trimming, and the 64,000 token output ceiling allows long generated artifacts. Reasoning support is useful for tasks where you want the model's intermediate thinking surfaced, for example when auditing how a fix was derived.

You call it through api14's OpenAI-compatible API using the model id cohere/north-mini-code:free, which means existing OpenAI SDK clients and request shapes work with only the base URL and model name changed. The request parameters it supports are frequency_penalty, include_reasoning, max_tokens, presence_penalty, reasoning, seed, stop, temperature, tool_choice, tools, top_k and top_p. In practice you set temperature, top_p or top_k to control sampling, use max_tokens to bound the response, use stop to end generation on a chosen sequence, and use seed for reproducible sampling. For agentic flows, pass tools and tool_choice, and use reasoning or include_reasoning when you want reasoning content returned.

Keep the input text-only, since no image or audio input is supported, and treat the free variant as the entry point for evaluation rather than a guarantee of any particular throughput. If your task is not code or agent oriented, or if you need modalities other than text, a different model on api14 may fit better.