Meta: Muse Glimmer 30B

Meta: Muse Glimmer 30B is a dense, open-weight multimodal model from Meta, developed by Meta Superintelligence Labs. It is distilled from Muse Spark and optimized for autonomous agents on consumer hardware. It sits in the text category and is positioned for long-horizon work, which makes it a candidate for agent-style applications rather than one-shot prompt and response tasks.

The model accepts text and image inputs and produces text output. Its context window is 131072 tokens, and the maximum output length is 16384 tokens. It supports reasoning, exposed through the include_reasoning and reasoning parameters, along with reasoning_effort for controlling how much reasoning work the model performs. Because it handles images as well as text, it can take screenshots, diagrams or other visual material as part of a request, though everything it returns is text.

Those capabilities point to a few practical uses. Long-horizon agent workflows are the clearest fit, given the combination of a large context window, tool calling and reasoning controls. Tool use is supported through the tools and tool_choice parameters, so the model can be wired into function-calling loops. Structured outputs and response_format let you constrain replies to a schema, which suits extraction, classification and data pipelines. Image input extends this to tasks where the model needs to read a screenshot or a document image before acting. The 16384 token output ceiling is worth keeping in mind when planning multi-step agent runs, since long reasoning traces and tool results consume that budget.

You call the model through api14's OpenAI-compatible API using the model id "meta/muse-glimmer-30b". The supported request parameters are frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k and top_p. Sampling can be tuned with temperature, top_p, top_k and min_p, while seed, stop, logit_bias and the penalty parameters give you finer control over determinism and repetition. Reasoning behaviour is configured separately through the reasoning and reasoning_effort parameters.

There is no stated limitation beyond the output cap and the text-only output modality. If your task needs audio or video input, or output in a non-text format, this model does not provide it. For straightforward short prompts that do not need tools, images or reasoning, a smaller or simpler model may be a more economical choice, but the facts here do not identify a specific alternative.