Meta Muse New Terms Explained: Muse Spark 1.3, Contemplating Mode, Muse Glimmer & More

Muse has only been public since September 8, 2026, and the launch week already produced a small dictionary: Muse Spark 1.3, Contemplating mode, Muse Code, Muse Glimmer, Muse Voice Transcribe, the Meta Model API — plus older lookalikes like Muse Image that keep tripping people up in search. If you have seen these names in headlines or app reviews and weren't sure what each one is, this guide sorts them out. For the agent itself, start with our what Muse is and how it works overview.

The quick glossary

Term	What it actually is
Muse	Meta's personal AI agent — does tasks for you (email, travel, shopping)
Muse Spark 1.3	The current model family version powering the Muse agent
Muse Spark 1.1	An earlier multimodal reasoning model, offered through the Meta Model API
Muse Spark 1.2	A version community users report using for coding; no standalone Meta announcement found
Contemplating mode	A Muse Spark mode that runs multiple agents in parallel and merges the results
Muse Code	Meta's developer-facing coding agent
Muse Glimmer	Meta's open, 30B-parameter model you can run locally (Apache 2.0)
Muse Voice Transcribe	Meta's real-time speech transcription model
Meta Model API	Meta's developer API for its models — not a way to control the Muse agent
Muse Image	An unrelated (and discontinued) Meta image generator from July 2026

The Muse Spark family: 1.1, 1.2, and 1.3

Muse the agent runs on the Muse Spark model family from Meta Superintelligence Labs. The version numbering is where most confusion starts:

• Muse Spark 1.3 is the current version — Meta published it on September 2, 2026, less than a week before the agent launched. When a review says "Muse is smart" or "Muse made a mistake," you are seeing 1.3 at work.
• Muse Spark 1.1 is an earlier release aimed at developers: Meta describes it as a multimodal reasoning model built for agentic tasks, with gains in tool use, computer use, and coding. Meta AI chief Alexandr Wang said it is "trained to know when to automate and when to use interfaces directly — it scripts when that's faster, clicks when direct."
• Muse Spark 1.2 sits awkwardly in the middle. Community users describe coding with it and praise its price, and developer pages reference it — but we could not find a standalone Meta announcement, so treat details about 1.2 as community-reported rather than official.

For everyday use the practical takeaway is simple: you don't pick a Muse Spark version in the Muse app. You just use Muse; the version underneath changes as Meta ships updates. If you want the broader context of how the model and the agent fit together, our main Muse AI guide covers it.

Contemplating mode: many agents, one answer

Contemplating mode is the launch week's most-searched new term. Meta introduced it alongside Muse Spark as a mode that "orchestrates multiple agents that reason in parallel" on a hard prompt, then synthesizes their work into a single answer.

The numbers Meta reports are striking: with tools enabled, Muse Spark in Contemplating mode scores 58% on Humanity's Last Exam and 38% on Frontier Science Research, and Meta says it outperformed all models tested on HealthBench. Community reports describe around 16 parallel agents per prompt; Meta's own post says only "multiple agents," so treat the exact count as unconfirmed.

If that sounds like OpenAI's high-compute "Pro" answers or Google's Deep Think, that's the right mental model — Meta positioned Contemplating mode against exactly those products. Our Muse vs ChatGPT and Muse vs Gemini comparisons show how the surrounding products stack up.

Muse Glimmer: the Muse model you can actually download

Muse Glimmer answers a question power users asked the moment Muse launched: can any of this run locally? For the main Spark models, no. But Meta also publishes Muse Glimmer — a 30-billion-parameter model, licensed under Apache 2.0, sized to run on a single GPU. Meta describes it as built for "always-on local agents," tuned for tool use and long-context work, with a perception encoder distilled from Muse Spark. The weights are on Hugging Face and LM Studio.

Glimmer matters for two audiences: developers building their own small agents, and privacy-minded users who like the idea of an agent that never leaves their machine. It is not the Muse agent — you don't get Muse's secure VM, approvals, or WhatsApp integration. If local privacy is your priority, it's also worth reading how the cloud-based Muse handles data in our Muse privacy and security guide.

Muse Voice Transcribe and Muse Code

Two more launch-window names belong to Meta's developer push:

• Muse Voice Transcribe is Meta's first real-time audio perception model: streaming speech recognition with speaker attribution (diarization) across 70+ languages, designed to transcribe live audio as it happens rather than after the fact. It's available through the Meta Model API and ships inside Meta's own tools.
• Muse Code is Meta's coding agent for developers — the same "does the work, not just suggests it" idea applied to programming tasks. It shares naming and model lineage with Muse but is a separate product aimed at engineers, not a feature inside the Muse app.

The Meta Model API — an API for the models, not for Muse

One release note that matters if you saw "Muse API" trending: Meta now runs a Meta Model API (at developer.meta.com), where developers can call Muse Spark models — and the models are also reachable through OpenRouter. What does not exist is an API for the Muse agent itself. Nobody outside Meta can programmatically drive Muse to send your email or book a trip; that only happens through the official surfaces: the Muse app, muse.ai, and WhatsApp. Anything advertising "Muse agent API access" is misinformed at best.

Two lookalikes that keep confusing people

Muse Image. Search suggestions still surface "Muse image" — but that's Meta's July 2026 image generator, not the agent. Muse Image went viral for letting users modify photos from public Instagram accounts, drew backlash from privacy advocates and Hollywood agencies, and Meta removed it on July 10, 2026 — two days after launch — saying it "missed the mark." Launch-week coverage of the agent has recycled the name, so check which Muse an article means.

Muse vs Siri. Meta timed the agent's launch for the day before Apple's September 9 event, and the press framed the week as Muse against Apple's newly overhauled Siri. As a search term, "muse vs siri" is really a question about two strategies: Meta's cloud agent that acts on your accounts versus Apple's on-device assistant overhaul. We'll cover it properly once both products' post-launch behavior settles.

Which of these new terms actually matters?

If you use Muse the agent: only Muse Spark 1.3 (what's under the hood) and Contemplating mode (how hard problems get answered) change your day-to-day experience. If you build things: the Meta Model API, Muse Code, and Muse Glimmer are the entries to watch. And if a term doesn't appear in the table above, be careful — Meta's "Muse" family now spans an agent, models, and discontinued side projects, and our guides keep track of which is which.

Keep reading

• What Is Meta Muse and How Does It Work?
• Muse vs Meta AI: the difference explained
• Free Muse prompt generator