IBM: Granite 4.2 8B

IBM: Granite 4.2 8B is a dense reasoning model from IBM, released under the model id ibm-granite/granite-4.2-8b. It belongs to the text category, so it is built for language work rather than image, audio or video processing. IBM positions it for mathematics, code generation, multilingual dialogue, and agentic workflows that need multi-step reasoning, which tells you the intended audience is developers who want a general purpose text model with reasoning behaviour rather than a narrow single-task system.

On the input side the model accepts text only and returns text only. It has a context window of 131,072 tokens, and the maximum output is 117,964 tokens, so long inputs and long generations are both feasible within the same request. The supported parameters include the usual sampling controls (temperature, top_p, top_k, min_p, frequency_penalty, presence_penalty, repetition_penalty, seed, stop, logit_bias, logprobs, top_logprobs) alongside reasoning-specific options such as reasoning, reasoning_effort and include_reasoning. It also supports structured_outputs and response_format, plus tools and tool_choice for function calling.

Those capabilities point to a few practical uses. Mathematics and code generation fit naturally, since the model is described as a reasoning model and can be asked to work through problems step by step. Multilingual dialogue makes it a candidate for chat assistants that serve users in more than one language. Because it supports tools and tool_choice, it can sit inside agentic workflows where the model decides which function to call and then continues from the result, including multi-step chains. Structured outputs and response_format help when you need JSON or another schema-shaped reply that downstream code can parse without extra cleanup.

Calling it through api14 works the same way as any other OpenAI-compatible model. Point your client at the api14 base URL, send a request to the chat completions endpoint, and set the model field to ibm-granite/granite-4.2-8b. You can pass any of the parameters listed above in the request body, including max_tokens to cap the reply, temperature and the other sampling knobs to control randomness, seed for reproducibility, stop sequences, and tools or tool_choice when you want function calling. Reasoning behaviour is controlled through reasoning, reasoning_effort and include_reasoning, and logprobs with top_logprobs are available if you need token-level probabilities.

The main constraint to keep in mind is modality: this is a text-in, text-out model, so tasks that require image or audio understanding need a different model. If your workload is purely text and benefits from reasoning, tool use or structured output, Granite 4.2 8B is a reasonable fit, and the 131,072 token context gives you room for large prompts and long answers.