AgentsSeptember 3, 2026via MarkTechPost
Meta AI Released Muse Spark 1.3: An Agentic Coding Model That Uses ~20% Fewer Tool Calls and ~25% Fewer Tokens Than Muse Spark 1.2
Why it matters
Two concrete agent-efficiency moves: Meta's coding agent uses fewer tool calls and tokens (measurable improvement in agent overhead); Perplexity's hybrid-compute architecture solves a real production constraint — splitting agent reasoning between cloud and device to preserve privacy without breaking context or autonomy.
Key signals
- Meta Muse Spark 1.3: ~20% fewer tool calls, ~25% fewer tokens vs. 1.2
- Perplexity hybrid compute: cloud (search, planning, reasoning) + on-device (sensitive steps)
- Context and task continuity maintained across cloud-to-device handoff
- On-device privacy classifier (0.6B) open-sourced; 0.629 character F1 score, highest among 12 evaluated detectors
- Perplexity hybrid compute live for Pro/Max/Enterprise on Apple silicon Macs with 24GB+ unified memory
- Privacy gate logic: keep local, mask, refuse, or ask for consent
The hook
Meta's Muse Spark 1.3 cuts tool calls by a fifth. Perplexity ships hybrid compute that keeps your data off the cloud.
Perplexity has shipped hybrid compute for its Mac app, splitting a single Perplexity Computer task between frontier models in the cloud and a compact model running on the user's machine. Tasks start in the cloud for search, planning and reasoning, then hand sensitive steps down to the Mac without re…