Muse Glimmer 30B vs Muse Spark 1.3

Muse Glimmer 30B is a dense vision language model from Meta built for long-horizon agentic work on local hardware. The architecture pairs a 52-layer causal text decoder with a roughly 1.8B parameter ViT-G/14 perception encoder for about 29.6 billion parameters in total, and it accepts interleaved text and image input so an agent can interpret screenshots, charts, and documents alongside conversation. The decoder uses grouped-query attention with 32 query heads and 2 key-value heads, a repeating pattern of three sliding-window local attention layers followed by one global layer, SwiGLU feed-forward blocks, and rotary position embeddings applied on the local layers, supporting a trained context of 131,072 tokens.

Meta describes the model as distilled from the larger Muse Spark and trained and evaluated around agentic behavior: end-to-end task completion, schema-accurate tool calling, multi-step reasoning across extended workflows, and recovery when a tool call returns an unexpected result. Reasoning effort is selectable across low, medium, high, and xhigh settings, and the model emits channel-scoped reasoning traces together with XML style tool calls rather than JSON, which requires parsers specific to this family. A companion block-diffusion drafter head predicts blocks of 16 tokens per forward pass for speculative decoding, with the main model verifying the proposals in parallel.

Muse Spark 1.3 is a proprietary multimodal reasoning model from Meta Superintelligence Labs and the fourth Muse Spark release in five months, arriving on September 2, 2026. It takes text, images, video, and document files as input and returns text, and it operates over a context window of 1,048,576 tokens. Meta trains the model for long-horizon agentic work, so it carries accumulated context and prior tool results forward across many turns, reconciles messy or conflicting inputs, and asks for clarification when a task is underspecified. Visual inputs such as screenshots and video clips feed a reasoning loop that runs against a real execution environment rather than a scripted sequence of steps.

The model exposes graded reasoning effort settings. An xhigh configuration is generally available at launch, while a max reasoning configuration aimed at harder reasoning and agentic problems arrives after further safety testing. Artificial Analysis measures Muse Spark 1.3 (max) at 62 on its Intelligence Index and the xhigh configuration at 61, with agentic tool-use evaluations driving most of the gain over Muse Spark 1.2; max reaches 52% on Tau3-Bench Banking by spending more turns and reasoning tokens than xhigh. Prior Muse Spark versions emit bounding box coordinates, transcriptions, and structured field extractions from images on Roboflow Vision Evals.