Urgent.News

What's breaking now, across thousands of outlets.

Tech

Meta's Muse Spark Open: 816B Parameters You Can Download Today (But Probably Can't Run)

Meta's Muse Spark Open: 816B Parameters You Can Download Today (But Probably Can't Run) Meta launched Muse Spark Open on October 15, 2026 — an open-weight model it calls the "most capable U.S. open-weight model to date." The weights are free to download right now. The catch: running it locally needs about a terabyte of RAM. Here's what developers need to know. What is Muse Spark Open? Muse Spark…

On October 15, 2026, Meta unveiled Muse Spark Open, an open-weight model boasting an impressive 816 billion parameters. This marks the most powerful U.S. open-weight model currently available. The model's weights are freely downloadable, but running it locally requires an astronomical 1 terabyte of RAM, effectively barring most developers from utilizing it directly.

Muse Spark Open is the open counterpart to Meta AI's Muse Wonder, launched just three days prior. The model is multimodal, capable of processing text, images, audio, and video, and is specifically designed for agentic coding tasks that involve extended tool use and bug fixing. Meta offers three avenues to access Muse Spark Open: downloading the free weights, running it locally using specialized hardware, or leveraging hosted APIs provided by partners such as Together AI, Fireworks AI, Groq, Modal, Nebius, and OpenRouter.

While the model shows promise in long-horizon agentic coding and multimodal understanding, the hardware requirements and licensing terms are significant barriers for widespread adoption.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Also reported by 1 other outlet

Read the original at dev.to →

More in Tech

Connected steps are not a workflow

Take a task that crosses two apps. A customer asks for a refund. The request lives in the support desk. The order lives in the billing system.

  • Refund request exists in support desk, order in billing system
  • Refunds over $50 require manager approval, not defined in API
  • No confirmation of refund processing before customer note sent

BERT's tokenizer bottleneck was hiding before WordPiece

BERT's tokenizer bottleneck was hiding before WordPiece Normalization consumed 88% of a warm English encoding pass with bert-base-uncased . WordPiece, the algorithm that chooses tokens, took 8%.

  • BERT tokenizer bottleneck hidden before WordPiece
  • WordPiece accounted for 88% of warm English encoding pass
  • Proposed branch improved BERT encoding 50x

More from Friday 16 October →