IBM: Granite 4.0 Micro

IBM Granite 4.0 Micro, listed on api14 as ibm-granite/granite-4.0-h-micro, is a text model from IBM's Granite 4 family. It is a 3B parameter model, and IBM describes the Granite 4 family as the latest in a series of models the company has released. This particular variant is the H-Micro member of that family, and IBM states that the family is fine-tuned for long tasks, which makes the Micro size an accessible entry point into the Granite 4 line for teams that want IBM's newer generation without moving to a larger model.

The model works with text input and produces text output only; it does not accept images or audio, and it does not generate anything other than text. Its context window is 131,000 tokens, and the maximum output is 117,900 tokens. That combination of a large context and a very large output allowance means you can send long documents in a single request and still leave substantial room for the model's reply, which is unusual for a model of this size.

The practical uses follow from those numbers. Summarising long reports, contracts or transcripts fits comfortably inside the context window. Extraction and structured generation are a natural fit too, since the model supports response_format and a long output ceiling lets it emit large JSON objects or multi-section documents in one call. Because the output limit is generous relative to the input window, tasks where the answer is itself lengthy, such as rewriting, expansion or drafting long passages from source material, are viable. General chat, classification, tagging and lightweight agents over text data are also reasonable starting points.

You call it through api14's OpenAI-compatible API using the model id ibm-granite/granite-4.0-h-micro, so existing OpenAI SDKs and clients work with only a base URL and key change. The supported request parameters are frequency_penalty, logit_bias, logprobs, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, temperature, top_k, top_logprobs and top_p. That set covers sampling control (temperature, top_p, top_k, min_p), repetition handling (frequency_penalty, presence_penalty, repetition_penalty), deterministic retries via seed, structured output via response_format, and token-level inspection through logprobs and top_logprobs.

There is no stated limitation beyond text-only input and output, and it is worth noting it is a 3B parameter model, so if you need vision, audio or a much larger model, a different option on api14 may suit better. Otherwise, for long-context text work with a high output ceiling, it is worth testing directly.