Meta's Muse Spark Open Achieves High Efficiency on NVIDIA GPUs

1 reports archived0 independent sourcesupdated 10/24/2026, 07:19:42

Synthesis & Latest Updates

Artificial Analysis released benchmarks for Meta's new open-weights model, Muse Spark Open. The MoE architecture (816B total, 42B active) demonstrates exceptional tokens-per-watt efficiency on NVIDIA GB300/B300 GPUs, leveraging sliding-window attention to significantly reduce KV cache requirements for long contexts.

LATEST/Artificial Analysis released benchmarks for Meta's new open-weights model, Muse Spark Open. The MoE architecture (816B total, 42B active) demonstrates exceptional tokens-per-watt efficiency on NVIDIA GB300/B300 GPUs, leveraging sliding-window attention to significantly reduce KV cache requirements for long contexts.

Hourly heat curve

128 fully observed hours

Now

0.0

Peak

22.608:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Artificial Analysis (@ArtificialAnlys)T1Oct 24, 07:19·85 pts
Meta's Muse Spark Open Achieves High Efficiency on NVIDIA GPUs
- Muse Spark Open uses an MoE architecture with 816B total parameters but only 42B active, balancing capability and efficiency.
- Achieves ~150M tokens/s per MW throughput on NVIDIA GB300/B300 GPUs.
- Sliding-window attention in 46 layers reduces KV cache requirements by ~75% at 1M context length.
View Related Reports & Analysis Original Source