Meta: Muse Spark 1.3 Contributor

Meta: Muse Spark 1.3 Contributor is a text-focused multimodal reasoning model from Meta. It sits in the contributor tier of the Muse Spark 1.3 line, which Meta describes as the cost-efficient tier aimed at experimentation, learning, and early-stage agentic, multi-agent, and coding workflows. The provider release date is 2026-09-02. If you are evaluating models for a project that needs to reason over mixed media rather than plain text alone, this is the entry point Meta offers in that family.

The model accepts text, image, video, and file inputs, and it returns text. Its context window is 1,048,576 tokens, and the maximum output is 943,718 tokens. That combination matters for workloads where the input is large and the expected answer is also long, such as summarizing a long recording, extracting structured data from a stack of documents, or holding a lengthy working state across a multi-step task. Reasoning behavior is exposed rather than hidden: the supported parameters include reasoning, reasoning_effort, and include_reasoning, so you can control how much internal deliberation the model does and whether that deliberation is returned to you.

Those capabilities point to a few practical uses. Document and media analysis is a natural fit, since you can pass files, images, or video alongside a text prompt and ask for a written result. Coding and agentic workflows are another, helped by tool calling and structured outputs, which let the model invoke functions and return responses in a schema you define. Multi-agent setups can use the same features to pass structured messages between agents. Because the model is positioned for experimentation and early-stage work, it also suits prototyping a pipeline before committing to a heavier model.

You call it through api14's OpenAI-compatible API using the model id "meta/muse-spark-1.3-contributor". The supported request parameters are include_reasoning, max_tokens, reasoning, reasoning_effort, repetition_penalty, response_format, structured_outputs, temperature, tool_choice, tools, top_k, and top_p. In practice you set the model field to that id, send your messages, and add whichever of these parameters your workload needs, for example tools and tool_choice for function calling, or response_format and structured_outputs when you need a predictable shape back.

One limit to keep in mind is that output is text only, so tasks that require generating images, audio, or video are out of scope here. The contributor tier is also described as cost-efficient and aimed at experimentation, learning, and early-stage workflows, which suggests that production workloads with the strictest quality demands may be better served by a higher tier in the same family. If your inputs are text only and your context needs are modest, a smaller text model may be simpler to operate.