Meta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window
THE SO WHAT
A 42B active-parameter, 1M-token open-weight model tuned for agentic use on Apple desktops compresses the gap between local and cloud assistants. If you're building agents, you now have a credible on-device or hybrid path—revisit your dependency on single-vendor hosted LLMs.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMeta and Apple team up to make Muse AI models run better on Macs
Meta’s claim that Muse Spark Open can run on dual M5 Ultra Mac Studios via Apple’s MLX framework shows on-device and desk-side training are moving into serious territory, not just demos. If you’re building internal tools, start segmenting workloads into what can live on local high-end machines versus what truly needs cloud-scale clusters.
An AI CEO is calling out the people who bash AI but can't stop using it
The tension between public AI backlash and private dependence is widening—political and community resistance won’t translate into real usage cuts anytime soon. For operators, that means plan for more permitting and narrative friction around data centers while assuming steady or rising internal AI demand.
Applied AIMeta opens a weaker AI model while keeping the frontier closed
The split between open Muse Spark Open and closed Muse Wonder is the new normal—open weights for the mid-tier, tight control at the frontier, aligned with EU AI Act language. If your roadmap assumes full open access to best-in-class models, recalibrate toward hybrid stacks and contractual access, not pure open-source.
Applied AIMemo: Accenture has told many employees, some of whom had been spending as much as $1,000 monthly, that it will now only spend $29 a month per person on Claude
A hard cap from $1,000 to $29 per person on Claude spend shows the enterprise AI bill is now under procurement discipline—shadow AI usage is getting pulled into budget lines. If you’re an internal leader, expect to justify AI tool ROI with real metrics, and if you’re a vendor, assume per-seat price compression and push for usage-based or outcome-linked models.