Meta has released Muse Glimmer, a 30-billion-parameter agent model that can run on consumer hardware after its weights are compressed. According to The Decoder, the compressed model requires less than 20 GB of memory, putting it within reach of systems below the hardware typically associated with large AI models.
Muse Glimmer is the first open model from Meta’s new Superintelligence Labs. In an accompanying essay, CEO Mark Zuckerberg outlines an aggressive strategy centered on out-copying China and selling compute capacity through auctions, although the available account does not provide the plan’s full operational details.
The release marks Meta’s return to open models, but the available account does not specify Muse Glimmer’s performance, license, or release timing for the compressed version.
Why it matters: A compressed 30B model requiring less than 20 GB of memory gives developers using consumer hardware a way to run Meta’s agent model without the high-end systems typically associated with large AI models, while Zuckerberg’s essay connects that distribution strategy to compute auctions.
OpenAI is expanding its Daybreak cybersecurity defense program with GPT-5.6-Cyber, a cyber-trained model designed to give defenders a head start on vulnerabilities before attackers find them. According to OpenAI, the model answers up to 98.5 percent of security queries that would otherwise be blocked and has already uncovered two previously unknown Chrome vulnerabilities.
The launch adds a named model to Daybreak’s effort to help cybersecurity defenders find flaws before attackers exploit them. OpenAI’s reported Chrome discoveries put vulnerability research at the center of the model’s initial pitch.
Why it matters: GPT-5.6-Cyber gives cybersecurity defenders a reported 98.5 percent answer rate on previously blocked security queries, while its two Chrome findings show the program targeting vulnerabilities before attackers exploit them.
Novo Nordisk is expanding its use of AWS AI tools across drug discovery. Under the agreement, AWS will become Novo Nordisk's preferred cloud provider and strategic AI partner, with AI agents supporting target identification, therapy design and research workflows.
The companies are also creating a co-innovation hub at Novo Nordisk's existing London site. The announcement does not disclose a deployment timeline, investment amount, specific AWS models or performance metrics, so the commercial impact remains unquantified.
Why it matters: For Novo Nordisk's R&D teams, AWS will support target identification, therapy design and research workflows, but the companies have not disclosed metrics showing faster discovery or better candidate selection.