Muse Spark
Meta·
LLMsflagshipcloud
Overview
First model from Meta Superintelligence Labs under Alexandr Wang. Powers Meta AI across Facebook, Instagram, WhatsApp, Messenger and Ray-Ban glasses. Natively multimodal (text/image/voice input, text output) with multi-agent 'Contemplating mode'. Closed-source departure from the Llama lineage. Free on meta.ai and the Meta AI app; API in private preview. Consumer-focused, strong in health and multimodal; coding gap acknowledged by Meta.
Capabilities and innovations
Multimodal perceptionVoice inputReasoning and agentic tasksPowers Meta AI productsRay-Ban smart glasses integrationHealth Q&AMeta Superintelligence Labs DebutContemplating Mode (multi-agent)Closed-source Departure from Llama
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 89.65% | independent GPQA Diamond 89,65%, measured by Vals AI. Artificial Analysis measures 88,38%. |
| Humanity’s Last Exam (no tools) | 39.9% | third party HLE 39,9% no-tools, measured by Artificial Analysis (39,94% raw). |
| MMLU-Pro | 87.32% | independent MMLU Pro 87,32%, measured by Vals AI. |
| MMMU-Pro | 80.5% | unsourced |
- SWE-bench Verified: not reported by the vendor
- SWE-bench Pro: not reported by the vendor
- MMLU: not reported by the vendor
- Terminal-Bench 2.x: not reported by the vendor
Links
More from Meta
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.