Meta Superintelligence Labs' 30-billion-parameter open-weight agentic model, released August 10, 2026 under the Apache 2.0 license. It pairs a 28B text decoder with a ~2B ViT-style vision encoder, taking interleaved text and images as input and producing text. Distilled from the larger Muse Spark 1.2, it runs on a single 24GB consumer GPU via 4-bit quantization. It supports a 131,072-token context, 100+ languages, and low/medium/high/xhigh reasoning settings, with day-0 runtimes across llama.cpp, MLX, Ollama, vLLM, and SGLang. Reported scores include 75.5 on MCP Atlas, 76.0 on SWE-Bench Verified, 51.7 on Terminal-Bench 2.1, 94.7 on AIME 2026, and 83.5 on GPQA Diamond.
Parameters
30B
Context Window
131K
License
Apache 2.0
Release Date
2026-08-10
API Pricing
API pricing for this model is not yet available
Strengths
- •Open weights that fit on a single 24GB consumer GPU at 30B
- •Apache 2.0 — Meta's most permissive open release since Llama
- •Strong agentic scores: 75.5 MCP Atlas, 76.0 SWE-Bench Verified
- •Text+image input lets it read screenshots, charts, and docs while calling tools
Weaknesses
- •Every benchmark figure is Meta self-reported, pending independent replication
- •Audio and video are not primary modalities — image input only
- •Teacher size and distillation recipe of Muse Spark 1.2 are undisclosed
- •Full precision exceeds 55GB; unusable without quantization on consumer hardware
Use Cases
- •Local, privacy-preserving coding agents
- •Automating internal docs and help desks that read screenshots and PDFs
- •Offline AI assistants running on consumer GPUs
- •Base model for fine-tuned agent scaffolds (OpenClaw, Hermes Agent)
Deep Analysis
Parameters
30B (28B text + 2B vision)
Dense causal transformer, 52 layers; ~1.8B ViT-G/14 perception encoder
Context Window
131,072 tokens
Knowledge cutoff January 4, 2026; 100+ languages
License
Apache 2.0
Meta's most permissive open release since Llama
VRAM to Run
24 GB (4-bit quantized)
Full precision >55GB; fits RTX 3090/4090 or Apple Silicon
MCP Atlas
75.5
Beats Gemma4-31B (54.2) and Qwen3.6-27B (62.5)
SWE-Bench Verified
76.0
Terminal-Bench 2.1 51.7; AIME 2026 94.7; GPQA Diamond 83.5
Strengths
- ・Open weights that run on a single 24GB consumer GPU after 4-bit quantization — local, private agentic inference without a hosted API.
- ・Apache 2.0, Meta's most permissive open license, enabling unrestricted commercial use, modification, and redistribution.
- ・Strong agentic scores (MCP Atlas 75.5, SWE-Bench Verified 76.0) and multimodal text+image input for screenshot, chart, and document understanding.
- ・Day-0 ecosystem support: llama.cpp, MLX, ExecuTorch, Ollama, LM Studio, vLLM, SGLang, plus a DFlash speculative-decoding drafter (3.1x on RTX 5090).
Weaknesses
- ・Every benchmark figure is Meta self-reported and still pending independent replication.
- ・Image input only — no audio or video as primary modalities.
- ・The Muse Spark 1.2 teacher's size and the distillation recipe are undisclosed.
- ・Full precision exceeds 55GB, so it is impractical without quantization on consumer hardware.
Competitor Comparison
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| Muse Glimmer 30B | 75.5 (MCP Atlas) | 76.0 (SWE-Bench Verified) | 83.5 (GPQA Diamond) | Apache 2.0; 24GB |
| Gemma4-31B | 54.2 (MCP Atlas) | 66.6 (SWE-Bench Verified) | 85.7 (GPQA Diamond) | Gemma license |
| Qwen3.6-27B | 62.5 (MCP Atlas) | 77.2 (SWE-Bench Verified) | 84.2 (GPQA Diamond) | Qwen license |
| Muse Spark 1.2 | 54 (AA Intelligence) | N/A | N/A | Closed; weights promised |
Muse Glimmer 30B is Meta Superintelligence Labs' open-weight agentic model, released August 10, 2026 under Apache 2.0. Distilled from Muse Spark 1.2, it runs locally on a single 24GB GPU, accepts interleaved text and images, and is tuned around the agent loop — plan, call tools, check results, recover from failure.
Sources
Analysis generated: 2026-09-02