← Dailies

Daily · 2026-10-16 00:00 UTC

Meta's 816B open-weight model runs on a Mac; FTC subpoenas OpenAI and Anthropic — 2026-10-16 00:00

Meta ships an 816B-parameter open-weight model that runs on Mac Studios, the FTC subpoenas OpenAI and Anthropic over safety incidents, Nvidia opens Nemotron 3.5, and more.

Source ↗

1. Meta's Muse Spark Open is an 816B open-weight model that runs on your Mac — Meta released Muse Spark Open, a natively multimodal open-weight model with 816 billion total parameters that Meta calls "the most capable US open-weight model to date." Meta says it's the first model in its class that can "run on your desktop using two Apple M5 Ultra Mac Studios with the latest MLX framework." Meta and Apple, usually rivals, worked together on the launch: Mark Zuckerberg posted that the weights drop today and "run great on new Mac Studios thanks to work with @johnternus and Apple," and new Apple CEO John Ternus replied, "Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk." The Verge · Oct 15
2. Anthropic's Q3 revenue hits $16.8B as its IPO looms — Anthropic told potential investors at a private meeting in San Francisco that third-quarter revenue hit $16.8 billion — roughly a 45% jump from Q2, and more than 3.6 times its total revenue for all of 2025. Growth has cooled: Q2 revenue had risen more than 140% on demand for Claude Code. The Information reports an adjusted operating margin of 11% for the year through Q3, and the Financial Times reports a second straight quarter of adjusted profitability, with $1.70 of revenue generated for every dollar spent on compute. The IPO is reportedly set for November, before Thanksgiving; Anthropic's private valuation jumped to $965 billion in May. The Decoder · Oct 15
3. Nvidia opens Nemotron 3.5 Super VL and two edge "Bolt" models — Nvidia released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model it says delivers the fastest task completion and highest accuracy for specialized always-on agents, alongside Nemotron 3.5 Bolt 74B-A9B 24GB for local agentic work on RTX/DGX Spark and Nemotron 3.5 Bolt E4B 4GB for fast conversational agents on GeForce RTX and Jetson. Super VL and Bolt 74B-A9B ship now; Bolt E4B arrives Oct 23. Nvidia says Super VL is 5x faster than the next best open agentic model in its class, and that Bolt 74B-A9B is the #1 open model in its class on PinchBench. NVIDIA · Oct 15
4. FTC subpoenas OpenAI and Anthropic over safety incidents — The Federal Trade Commission has subpoenaed Anthropic and OpenAI, an agency spokesperson confirmed, seeking documents on safety incidents, privacy policies and safeguards, according to Axios. The civil investigative demands went out Wednesday, and the agency is expected to issue subpoenas to third-party organizations at a later date; it could also compel the companies to testify before Congress. The FTC had said two weeks earlier that it was investigating both labs as part of a probe it launched this summer, after incidents in which their models "went rogue and hacked into websites." CBS News · Oct 15
5. AI-debt jitters deepen a global tech sell-off — An accelerating sell-off in AI-linked shares hit global markets as fears over rising inflation and future rate hikes spooked traders already wary of the sector's debt load. South Korea's Kospi 200 fell more than 8% overnight to a six-month closing low, triggering circuit breakers at points; SK Hynix and Samsung Electronics dropped nearly 11% and 9.5%. Japan's Nikkei 225 fell 4.9% as memory maker Kioxia slid 11% and SoftBank lost 10%. The falls followed a Wall Street slide — the S&P 500's worst since July — after US September CPI came in at 3.7% and producer prices rose 6.3% year on year. The Guardian · Oct 15
6. Online ads were weaponised to deliver zero-click spyware, Citizen Lab says — Researchers at the University of Toronto's Citizen Lab say an Israeli ad agency, Discover-Tech, served millions of online ads that let "zero-click" spyware be installed on targets — no link click or file download needed. Its Controlium platform, they say, can "mass-profile and deanonymise" internet users, then hand a selected target to a third-party spyware firm that can extract messages, photos and emails, record calls and secretly activate microphones. Citizen Lab believes dozens of devices at Al Jazeera were compromised; Discover-Tech has served ads in Yemen, Brazil, Mexico and Czechia since 2024, and related work by Confiant suggests networks tied to the UN, NATO and the ICC were exposed. The Guardian · Oct 15
7. Amazon's Bee launches two new personal AI devices, starting at $75 — Amazon's Bee introduced the Bee Frontier, a next-generation device that clips to the wrist or pins to clothes, and the credit-card-sized Bee Eclipse, which magnetically attaches to a phone. The Frontier costs $74.99 and the Eclipse $99.99, undercutting rivals such as the Plaud Note ($159) and Pocket ($129). The devices summarize conversations and turn them into to-dos; summaries and 14 days of memory are free, while an $11.99/month Bee Pro subscription adds unlimited summaries and integrations with Amazon.com, Alexa+ and third-party apps. Audio is processed in real time and deleted rather than stored. TechCrunch · Oct 15
8. Black Forest Labs drops FLUX 3 Image Klein as open weights — Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model for text-to-image generation, multi-reference generation and image editing that supports native 4K output, pixel-accurate edits and up to ten reference images. The company says it ranks first among open models in its internal evaluations and rates above closed rivals Ideogram V4.5 and FLUX.2 [pro]. The release includes a base model for fine-tuning plus FP8, INT8 and NVFP4 builds developed with Nvidia support; the generator weights need 11.2 GB in BF16 and 5.6 GB in FP8, making local and on-device editing practical. The weights ship under the FLUX Kommunity License, free for commercial use by companies making under $5 million a year. Black Forest Labs · Oct 15
9. Krea ships Krea 3, an image-editing model built for production — Krea released Krea 3, an image-editing model it bills as four times faster than Nano Banana Pro. Krea's own comparison puts an edit at 8 seconds versus 26.8 seconds for Nano Banana Pro and 39.9 seconds for GPT Image 2.5 Sunburst, measured as median request-to-result time across a 12-image batch including queue time. Krea says the model renders natively at 2K, supports regional editing that changes only the pixels you ask for, and keeps products and faces consistent across a sequence of edits. Krea · Oct 15
10. Bluesky's Attie now lets anyone build sites, games and quizzes — At TechCrunch Disrupt 2026, Bluesky expanded its experimental Attie platform so users can create websites, quizzes, games, polls and other interactive experiences — and publish them to the same AT Protocol (ATProto) that powers Bluesky. CEO Toni Schneider called it "social vibe coding or vibe coding for your community," arguing that today's solo vibe-coding experience should become a group one. Creations connect to a user's ATProto identity, can bring their own AI model or use open-weight options, and tap into Bluesky's wider audience. TechCrunch · Oct 15

🧪 Try This Next

• 🔧 Test Meta's Muse Spark Open locally — the first open-weight model marketed as runnable on two M5 Ultra Mac Studios through Apple's MLX, and the weights are out now.
• 🎥 Watch Krea's Krea 3 side-by-side edits — an 8-second edit next to Nano Banana Pro's 26.8 seconds is the clearest read yet on where production image editing is heading.
• 📝 Save the Citizen Lab findings on ad-delivered spyware — a new, interaction-free infection vector worth understanding if you touch anything ad-adjacent.

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.