Your signal. Your price.
Meta’s Muse Spark Open has 816 billion total parameters, 42 billion active parameters, and a million-token context window. Meta claims open-state-of-the-art results on Jobbench and MCP Atlas; Artificial Analysis scored it 39, behind Kimi K3 on coding tests but ahead of Nemotron 3 Ultra and Inkling.
Meta’s open-weight release restores a credible US alternative after Llama 4 disappointed and the company rebuilt its AI division. Nathan Lambert expects strong adoption if benchmark results hold; Box CEO Aaron Levie says open models let enterprises combine frontier orchestration with cheaper, tailored models for repeatable workflows.
Apple updated its MLX toolkit so Muse Spark Open runs on Mac hardware, despite years of tension between Apple and Meta. The full model requires a terabyte of memory and two high-end Mac Studios, putting private inference within reach of some startups; quantized versions can run on smaller machines with trade-offs.
Mariana van Zeller cites Reuters reporting that Meta internally estimated 15 billion scam ads appeared on its platforms daily and projected $16 billion from scam ads and banned goods. Meta reportedly removed advertisers at 95% fraud confidence but let lower-confidence suspected scammers buy ads at higher auction prices.
Companies are reversing rushed AI rollouts: Meta reportedly rebuilt management after AI errors, Microsoft employees complained about AI-generated documents, and Walmart workers faced false alerts and unreasonable expectations from AI tools.
Adam Curry says he uses at least eight coding agents and runs five local machines that do not send data to the cloud. He sees personal data files as useful agent context, while acknowledging that a virtual machine rented from Meta raises an ownership question.
Sam Sethi notes that some platforms block AI agents while Shopify is making its stores easier for Meta’s Muse agents to use for purchases. He says TrueFans has faced rising MongoDB costs from AI bots and uses a web application firewall informed by John Spurlock’s O.Porg list.
Ithaca Holdings, led by former Sequoia partner Ravi Gupta and former Meta executive John Hageman, announced a deal to take control of struggling Bumble. Gupta argues that accessible intelligence could make human connection more valuable and that a small, AI-focused team can rebuild a business; Bumble’s market capitalization peaked at $14 billion after its 2021 IPO before its share price fell 95% the following year.
Meta released Muse Wonder and Muse Wonderful after investing $14.3 billion for 49% of Scale AI and recruiting Alexander Wang to lead its AI lab. Salim sees active memory and Meta's distribution as advantages; Alex said Muse remained below the cost-performance frontier and questioned whether consumer use would justify the model spending.
Nathan Lambert says model progress depends on compute, data, and talent, plus the ability to organize them. OpenAI and Anthropic have made the earliest, most ambitious bets; Meta’s reset put it back on a positive trajectory.
Meta launched Muse Wonder and Wonderful, achieving a 56 score on the Artificial Analysis intelligence index and 79% on DeepSuite 1.1. At $4.52 per task, Wonder costs nearly 3x more than Muse Spark 1.3 while targeting agentic workflows.
Developers criticized Muse Wonder for failing 3D game tasks and scoring 52% on Dan Shipper senior engineer benchmark. Alexander Wang acknowledged omitting 3D training, emphasizing that Meta designed the models around active memory and focus for personal agents.
Mark Zuckerberg rejected the Pacing the Frontier safety agreement, maintaining that open releases with standard guardrails present lower systemic risks than unreleased internal models. Mark Zuckerberg noted Meta has reached the frontier and expects development velocity to accelerate.
Tristan Harris highlights how social media business models optimize for user engagement over mental health, driving algorithmic design toward doomscrolling. The resulting public harms catalyzed a $17 billion lawsuit against Meta after state attorneys general reviewed internal research.