Last updated September 18, 2026

Meta Muse Spark in Codex — Use Meta Models | Codex Router

Meta in Codex means adding Meta’s Muse Spark models to the Codex model picker through Codex Router. Muse Spark speaks the Responses protocol, so it fits the Codex transport directly without a separate adapter.

What Muse Spark is

The Muse Spark family is Meta’s Responses-protocol model line. The meta provider points at https://api.meta.ai/v1 and authenticates with a META_API_KEY stored through the router’s hidden prompt. Requests are only sent to Meta’s endpoint for turns that select a Meta model.

Enable Meta in Codex

Install Codex Router first (see the install guide), then store the key and enable the provider:

Create a Meta Model API key.
Store it with the hidden prompt.
Enable the provider and restart Codex.

./bin/model-router codex provider-key meta set
./bin/model-router codex providers enable meta

On Windows use ./model-router.ps1 codex with the same commands. Then fully quit Codex, reopen it, and create a new task.

Registry models

Five Muse Spark models ship in the registry:

Picker label	Model ID
Muse Spark 1.3	`meta/muse-spark-1.3`
Muse Spark 1.3 Contributor	`meta/muse-spark-1.3-contributor`
Muse Spark 1.2	`meta/muse-spark-1.2`
Muse Spark 1.2 Contributor	`meta/muse-spark-1.2-contributor`
Muse Spark 1.1	`meta/muse-spark-1.1`

Add more from the account’s live catalog with ./bin/curate-models meta. Set META_BASE_URL to point at a different endpoint when you need one.

Context and reasoning

The shipped Muse Spark models advertise a 1M context window. They expose a reasoning-effort ladder from minimal through xhigh, and reasoning summaries are enabled, so the picker reflects the effort you choose rather than a fixed default.

The Contributor tier

The Contributor variant is cheaper than the standard tier, but Meta may use Contributor inputs and outputs for training. That is a data-policy trade-off, not a billing trick: pick the standard tier when the content must stay out of training, and the Contributor tier when the lower price matters more than that boundary.

Meta’s direct Responses endpoint rejects a self-referencing tool schema with HTTP 400 Recursive JSON schemas are not currently supported, which would lose the whole turn before inference. The verified meta/muse-spark-1.3-contributor route now breaks only the cycle-closing reference edge through the toolSchemaRecursion: "flatten" repair, preserving definitions, acyclic references and sibling constraints. Other Meta routes keep their payloads unchanged until their own endpoint proves the same restriction. Fully quit and reopen Codex after updating so the repaired catalog is loaded.

Verify the integration

./bin/model-router codex doctor
./bin/model-router codex providers
./bin/test-model 'meta/muse-spark-1.3' --live --yes

The provider must report SHOW and ready. If entries are missing, see models missing from the picker.

Provider policy

Meta may restrict or block proxy traffic. Routing Muse Spark through Codex Router is at your own risk; check Meta’s terms before relying on it.

FAQ

What are Meta Muse Spark models in Codex? Muse Spark is Meta’s model family that speaks the Responses protocol. Codex Router’s meta provider publishes the Muse Spark registry models into the Codex picker.

What is the Muse Spark Contributor tier? The Contributor tier is a cheaper variant of Muse Spark. Meta may use its inputs and outputs for training, so choose it according to your data policy.

How much context do Muse Spark models support? The shipped Muse Spark models declare a 1M context window, with reasoning efforts from minimal to xhigh and reasoning summaries enabled.