What Meta’s new open-source AI model means for the future of artificial intelligence

Summary
This PBS NewsHour segment, anchored by Geoff Bennett and reported by William Brangham, covers Meta’s release of its open-weight agentic AI model, Muse Glimmer. Brangham discusses Mark Zuckerberg’s accompanying 14-page essay advocating open-source AI and interviews The New York Times Silicon Valley correspondent Mike Isaac on the safety and commercial debate between open-weights and closed-source competitors like OpenAI and Anthropic.

What is shown

• [00:36] Demo footage of Meta’s Muse Glimmer | Agentic Task Completion in OpenCode, showing the model planning and executing multi-step tasks from a single prompt configuring a Home Assistant dashboard on a Raspberry Pi.
• [00:47] On-screen graphic quoting Mark Zuckerberg’s essay on technology openness.
• [04:08] The resulting responsive A/V receiver web interface running locally, built by Muse Glimmer.
• [04:16] A performance comparison benchmark of Muse Glimmer – Speculative Decoding running via ExecuTorch on a MacBook Pro (M5 Max, 64GB Unified Memory), comparing 57.8 tok/s (DFlash enabled) against 26.2 tok/s (disabled).
• [05:54] B-roll of large-scale AI data center construction sites and developers using coding environments.
• [06:34] Infographic slide citing Pew Research Center survey data on American AI chatbot usage.

Claims & numbers

• William Brangham states Meta released a new free, open-source program named Muse Glimmer alongside a 14-page essay by Mark Zuckerberg.
• The on-screen benchmark indicates Muse Glimmer achieves 57.8 tokens/second with DFlash Speculative Decoding enabled versus 26.2 tokens/second without on a MacBook Pro M5 Max with 64GB unified memory [04:16].
• Mike Isaac notes that closed-source developers (such as OpenAI and Anthropic) argue advanced models present severe proliferation risks if released openly, including cyberattacks and biological weapon development.
• Mike Isaac claims Meta is investing billions of dollars in data centers and talent to avoid falling behind closed-model leaders.
• An end-screen graphic cites a Pew Research Center survey (February 17–23) finding that about half of Americans report having used an AI chatbot, up from one third in 2024 [06:34].

Notable quotes

• Mark Zuckerberg (quoted graphic, [00:47]): “...openness and putting the power of tech into more people’s hands, not fewer.”
• Mike Isaac [02:36]: “The folks of open source say, hey, more eyeballs equals safer, we can figure out different security issues.”
• Mike Isaac [05:50]: “Meta is very behind in this AI race, you know, they’ve spent billions of dollars developing data centers... and they’re still getting beaten by Anthropic and OpenAI.”

Assessment
This is a standard broadcast news report and analytical interview rather than a direct technical product demonstration. The operational clips of Muse Glimmer and its speculative decoding performance are pre-recorded screen captures provided by Meta rather than tested live on air.

Described by gemini-3.8-flash on 2026-10-12 from the video’s audio and frames.