← Model indexMODEL PROFILE / muse-spark-1-3

Meta / current

Muse Spark 1.3

Meta's recommended closed-weight flagship API tier while Wonder remains in preview.

Context1M
ModalityText · Image · Video · Audio
AccessAPI
ReasoningControllable

Where it fits.

Optimised for agentic workflows and coding with fewer tool calls and tokens than earlier Spark revisions at unchanged pricing. Shipped September 2, 2026 at $1.25 input and $4.25 output per million tokens with cheaper cached input, plus a low-cost contributor tier at $0.10 and $0.20 that permits Meta to train on prompts. It remains the model Meta recommends for new work while Wonder is preview.

02

Core capabilities

  • 1M-token context trained for agentic work
  • Around 20% fewer tool calls than Spark 1.2
  • Text, image, video and document input
  • All reasoning-effort levels including max
03

Best-fit workflows

  • Agentic workflows
  • API-first coding tools
  • Multimodal document analysis
04 / OPERATIONS

From evaluation
to production.

Deployment notes

Default API tier for new Meta integrations. Put non-negotiable rules first in the system prompt, set reasoning effort explicitly, use native tool definitions, and keep Sovereign data on Muse Spark Open instead of the contributor tier.

Watch before choosing

Audio input is degraded relative to Spark 1.2, with Meta routing speech to Muse Voice Transcribe instead. The contributor variant trades privacy for price; never send confidential data there. Audio paths regress versus Spark 1.2.

05

Family & versions

Specifications and positioning checked against provider documentation.

Open official source↗