Day-Zero Access to Meta’s New 30B Agentic Model on Actual (SN95)

Actual (SN95) added day-zero support for Muse Glimmer, Meta’s newly released 30B open-weight agentic model featuring tool use, multi-step reasoning, multimodal capability, and Apache 2.0 licensing.

Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats…

— Mark Zuckerberg (@finkd) August 10, 2026

The model is quantized to under 20GB with speculative decoding enabled, so it runs at real usable speed directly on personal hardware. No cloud provider between the user and the model, no rented GPU, no subscription.

Muse Glimmer sits in what looks increasingly like a golden model size for personal inference, and Actual made it usable from day one.

What Muse Glimmer Brings

Meta’s new release checks the boxes agentic workflows need out of the box.

1. 30B open-weight parameters: Enough capability to handle real tasks, small enough to run locally.

2. Tool use built in: No wrapper needed to plug into agent frameworks.

3. Multi-step reasoning: Handles chained tasks without breaking.

4. Multimodal input: Text and images processed natively.

5. Apache 2.0 licensing: Free to use, modify, and deploy commercially.

6. Quantized under 20GB: Fits comfortably on machines most builders already own.

7. Speculative decoding enabled: Real usable speed rather than the sluggish experience quantized models often produce.

The combination gives serious builders an open agentic model powerful enough to do real work without requiring datacenter-scale infrastructure.

Why Actual (SN95) Was the Right Landing Spot

Actual’s whole product surface is designed for exactly this use case.

1. Personal compute as the target: Actual treats the user’s own hardware as the primary surface, not an afterthought.

2. Low CPU impact during inference: Machines stay usable while running models.

3. Multi-platform portability: Move between machines without breaking anything.

4. Native Hermes Agent support already shipped: Anyone running Muse Glimmer through Actual can plug directly into the Nous Research harness.

Firing Up Hermes Agent on Your Own Compute with Actual (SN95)

The playbook is not to build another datacenter, connect computers that already exist, run the best open models on them, and let agents use the compute.

Day Zero Is the Whole Point

Day-zero support for a new frontier open model separates infrastructure worth using from infrastructure that only talks about being useful.

Muse Glimmer arriving on Actual the same day Meta released it means anyone curious about the model can try it immediately on hardware they already own, with no waiting for a hosted provider to spin up an endpoint.