OpenClaw Adds Muse Spark 1.2—But Contributor Pricing Changes Data Use

OpenClaw has merged two new Meta provider choices into main: muse-spark-1.2 and muse-spark-1.2-contributor.

The model checkpoint is the same Muse Spark 1.2 family, but the service tiers are not interchangeable. Meta’s discounted Contributor tier permits training use of submitted and generated content and expressly says users must not submit sensitive, confidential or personal information.

That makes model selection a data-governance decision, not merely a price/performance toggle.

What OpenClaw adds

PR #120373 adds both IDs to the Meta plugin catalog while retaining Muse Spark 1.1 as OpenClaw’s default. The new entries advertise the modalities OpenClaw’s model manifest can currently represent directly—text and image—rather than implying native support for every modality listed by Meta.

Meta’s current catalog lists a 1,048,576-token context window for Muse Spark 1.1, 1.2 and 1.2 Contributor. OpenClaw records that exact value and retains a 131,072-token output cap in its provider catalog. The provider wrapper supplies the catalog cap only when neither caller nor provider payload already supplied one.

The feature is merged source. Operators still need a published @openclaw/meta-provider package containing it before those IDs become available in an installed Gateway.

Standard and Contributor have different data terms

Meta’s official model and pricing pages distinguish:

• Standard: Muse Spark 1.1 and 1.2; Meta says prompts and completions are not used to train Meta models.
• Contributor: Muse Spark 1.2 Contributor; substantially lower token prices in exchange for permission to use prompts and completions to train future Meta models.

At verification time, Meta listed Standard pricing of $1.25 per million input tokens, $0.15 cached input and $4.25 output. Contributor was listed at $0.10 input, $0.002 cached input and $0.20 output. Rates can change; check Meta’s live page before a cost decision.

The legal boundary is more important than the discount. Section 6.2 of Meta’s current Model API terms says users must not submit sensitive, confidential or personal information to Discounted Services. It also places responsibility for end-user location, disclosures and eligible traffic on the operator.

OpenClaw does not add a second consent checkbox. Selecting and using the Contributor model is the consequential action under Meta’s terms.

Recommendation

Use Standard Muse Spark 1.2 for ordinary production workloads unless an approved data-classification policy explicitly permits Contributor processing.

Do not route these classes to Contributor:

• proprietary source code or unreleased product plans;
• customer records, support transcripts or personal data;
• credentials, access tokens or security incident material;
• confidential contracts, finances or internal communications; or
• end-user traffic whose location or consent status has not been established.

A prompt scrubber is not enough if tools can fetch confidential workspace content after the initial request. Model policy must cover the whole agent turn: prompt, files, retrieval, tool results and generated output.

Rollout checklist

1. verify the installed Meta provider package contains both 1.2 IDs;
2. keep 1.1 or Standard 1.2 as the default during evaluation;
3. review Meta’s current terms, geographic policy and live prices with whoever owns privacy and contractual compliance;
4. separate Contributor into a profile or policy lane that cannot read confidential workspaces or production tools;
5. use synthetic evaluation data and confirm logs identify the exact model ID and tier;
6. test that policy cannot silently fall over from Standard to Contributor during provider failover; and
7. recheck terms and model designation periodically—the current terms describe a limited preview and may change.

Evidence and limits

The merged PR reports live request proof for the output cap and image input on all three model IDs. OpenClaw Academy verified Meta’s public model, pricing and terms pages, but did not send paid requests or independently benchmark model quality.

A successful color-recognition fixture proves wire compatibility, not general multimodal performance. The million-token catalog window is provider capacity, not a recommendation to fill it on every turn.

Bottom line

Muse Spark 1.2 broadens the Meta provider’s useful choices. The Contributor discount is real, but so is the data-use trade. Keep Standard as the production default; make Contributor an explicitly governed lane with no sensitive, confidential or personal information.