How do I switch a Hermes agent to Muse Spark 1.2?

Run hermes inference set openrouter and hermes model set meta/muse-spark-1.2, or set inference.provider and model.default in /opt/data/config.yaml. On managed instances use /model meta/muse-spark-1.2 from chat.

← Home

Guide

Hermes Agent + Muse Spark 1.2: Setup via OpenRouter

Q: Does Hermes Agent support Muse Spark 1.2?

Yes, through OpenRouter. Hermes treats OpenRouter as its default aggregator, so set the provider to openrouter and the model to meta/muse-spark-1.2. No Muse-specific plugin is needed.

Q: Can I run Muse Spark 1.2 locally with Hermes?

No. Muse Spark 1.2 weights are proprietary and unpublished, so there is no self-hosted route. Use an open-weights model such as one served by Ollama for local inference.

Meta released Muse Spark 1.2 on 5 August 2026 with a 1M-token context window and agent-focused training. Because Hermes Agent already treats OpenRouter as its default aggregator, switching a Hermes bot onto it is a two-command change. Here is the setup, plus the one tier you should not click without reading.

Why Muse Spark Suits a Hermes Agent

Hermes runs long. A persistent agent accumulates conversation history, tool output, and file context across days, and the usual failure mode is a context window filling up until the agent starts forgetting the beginning of its own task. A 1M-token window pushes that ceiling far enough out that most real workloads never touch it.

Muse Spark 1.2 also accepts text, images, video, audio, and PDF as input and returns text. For a Hermes bot fielding documents and screenshots on a messaging platform, that removes a conversion step.

Model ID and Pricing

Property	Value
Model ID	`meta/muse-spark-1.2`
Context window	1M tokens
Input modalities	Text, image, video, audio, PDF → text out
Standard price	$1.25 / 1M input, $4.25 / 1M output
Cached input	$0.15 / 1M
Benchmarks (Meta-published)	82.9% Terminal-Bench 2.1, 59.3% DeepSWE 1.1

Verify current pricing at openrouter.ai/meta/muse-spark-1.2. For the fuller model write-up — benchmark context and the availability caveat — see the OpenClaw Muse Spark guide.

Skip the contributor tier on a shared bot. Meta sells muse-spark-1.2-contributor at $0.10 input / $0.20 output in exchange for permission to train on your prompts and completions. A Hermes agent connected to a team chat is reading messages that belong to other people, and they did not agree to that. Use the standard tier unless every input is your own.

Option 1: OpenClaw Launch (Managed)

Managed Hermes instances route inference through OpenRouter already, so there is no key wiring:

Open the configurator at openclawlaunch.com and start a Hermes deploy.
Muse Spark is new, so it may not be a curated one-click pick yet — check the model list. If it is not there, add your own OpenRouter key under BYOK and enter meta/muse-spark-1.2 as the model. The Hermes BYOK guide covers this.
Connect a channel and click Deploy.

Switch at runtime from any connected channel with /model meta/muse-spark-1.2 — no redeploy.

Option 2: Self-Hosted Hermes Agent

Set your OpenRouter key, then point Hermes at the model:

export OPENROUTER_API_KEY=sk-or-...

hermes inference set openrouter
hermes model set meta/muse-spark-1.2

Or edit the config file directly:

# /opt/data/config.yaml
inference:
  provider: openrouter
model:
  default: meta/muse-spark-1.2

Get a key at openrouter.ai/keys. Full provider setup lives in the Hermes + OpenRouter guide, and a from-scratch install in the Hermes deploy guide.

One Provider, No Fallback

OpenRouter is Meta's partner for Muse Spark and forwards requests to a single upstream provider rather than spreading them across several. There is no second provider to absorb an outage. If your Hermes agent needs to answer at all hours, keep a fallback model configured and know which command switches to it.

Troubleshooting

Model not found. The ID is meta/muse-spark-1.2, not muse-spark-1.2 — OpenRouter IDs carry the vendor prefix.
402 from OpenRouter. The key has no credit, or a spend limit is capping it. Check your OpenRouter dashboard before assuming a Hermes problem.
Config edits do nothing. Hermes freezes defaults into config.yaml at first boot; edit that file rather than expecting a new default to apply to an existing instance.
Costs higher than expected. A 1M window is billed like any other context. Long sessions resend history each turn — the cached-input price is the number to watch, not the headline input price.

FAQ

Does Hermes Agent support Muse Spark 1.2 natively?

Through OpenRouter, yes — Hermes treats OpenRouter as its default aggregator, so any model OpenRouter lists is reachable by ID. There is no Muse-specific plugin to install.

Can I run Muse Spark 1.2 locally with Hermes?

No. The weights are proprietary and unpublished. For a local setup use an open-weights model instead — see Hermes + Ollama.

Which is better for Hermes: Muse Spark 1.2 or MiniMax M3?

Both carry 1M-token windows. MiniMax M3 is cheaper on input; Muse Spark 1.2 posts higher vendor-reported Terminal-Bench scores. For cost-sensitive long-context work start with M3; for agentic CLI and coding work, try Muse Spark and compare on your own tasks.