AI

Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full on Monday, its first attempt at a frontier model. By Meta's own testing they score comparably to Fable 5.1 and GPT-6 Astra, and Meta highlights "active memory", recall of what a user has told the model over weeks or months. Wonder costs $4 per million input tokens and $20 per million output tokens through Meta's API; Wonder Full is only for Muse Code subscribers.

On Thursday Meta also released the weights for Muse Spark Open, an 816-billion-parameter (42 billion active) multimodal model with a 1-million-token context window, which it calls the most capable US open-weight model so far. It is offered through hosting providers and on Hugging Face, and Meta says it runs on two Apple M5 Ultra Mac Studios.

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini, who previously developed the AlphaChip system, told TechCrunch Disrupt their startup is building AI that designs chips, learns from each iteration and improves the next design. Goldie said compute demand could grow at least 10,000-fold in five years and argued for more efficient chips over more data centers. They say they optimize whole chips rather than splitting them into dozens of blocks, and that a one-week design cycle would make custom chips for niches like AR, space and biological computing worthwhile. Results beyond AlphaChip have not yet been published.

At its first Config event in India, Figma announced Figma Specs, a Markdown editor where a team writes down its design language so that people and AI agents, Figma's own and third-party ones, can follow the same rules. Specs can reference canvas elements, design and code libraries and FigJam diagrams. Figma also opened its code layers feature to a public beta, added support for more than 20 tools through its MCP server, and is integrating the Payload headless CMS it acquired last year into Figma Make. Figma says India is its biggest market by users outside the US.

Deezer said it is building a toggle, due before the end of the year, that removes AI-generated tracks, albums and related artists from a listener's search results. The company says more than 50% of its daily uploads, about 90,000 tracks a day, were AI-generated as of July, and cites a survey in which about 45% of customers wanted a way to remove them.

YouTube Music took a disclosure approach: a new label on the player page, also shown in the main YouTube app, marks songs as Fully Gen AI, Partly Gen AI or No Gen AI, based on what partners declare and YouTube's own detection. AI involvement alone doesn't make a track ineligible for monetization on YouTube.

Australian startup Apate.ai, spun out of Macquarie University research, runs thousands of AI agents that pose as potential victims: they answer scam calls, reply to phishing messages and join scam groups, in many languages and voices. The company says its agents have wasted 17 years of scammers' time across 2.5 million conversations, and that the goal is to extract bank accounts, crypto wallets, phone numbers and impersonated brands, which it sells to banks, telcos, governments and utilities. Apate was a Startup Battlefield finalist at TechCrunch Disrupt; the article cites FBI figures of more than 22,000 AI-related crime complaints in 2025 with about $893 million in losses.

San Francisco startup Industrial Next, founded in 2021 by former Tesla engineers Lukas Pankau and Allen Pan, builds self-contained robotic stations driven by its own foundation models linking vision, task context and motion. It says a production line can be retooled in minutes instead of weeks without traditional programming. On the TechCrunch Disrupt Startup Battlefield stage the company announced it is working with BMW Group's Technology Office to evaluate its dexterous manipulation for manufacturing.

Google says Android users can now build an app with Google AI Studio and install it through Google Play for their own use only: it won't appear in search or listings and isn't shared with anyone else, though it must still follow Play's policies. The feature starts in the US for users 18 and over. Google says sharing with a small group of trusted people is coming in the following months, with a path to publish a personal app publicly after trust and safety review.

Santa Monica startup Blueprints AI, founded by architect and contractor Al Dram with Coco Cao, says its software turns a 2D sketch or PDF into code-compliant, permit-ready construction drawings in minutes. Dram says architects spend about 80% of their time drafting and 20% designing, and wants to flip that ratio. The company, part of TechCrunch Disrupt's Startup Battlefield 200, planned to unveil a second-generation version at the event. Claims about permit acceptance come from the company.

Munich-based CedarDB, a Technical University of Munich spinout built on the Umbra research engine, runs both transactional and analytical workloads in one database and is compatible with PostgreSQL, which CEO Moritz Sichert says keeps switching costs low and suits how LLMs write queries. After launching an on-premise enterprise product and a free community edition in 2025 alongside a €5.3 million seed round, the company announced limited availability of CedarDB Cloud at TechCrunch Disrupt.