Meta and Nvidia each released open-weight AI models this week, joining a broader U.S. industry push to compete with Chinese laboratories as the debate over open-source AI regulation intensifies in Washington.
Meta and Nvidia each released open-weight artificial intelligence models this week, entering a market where Chinese laboratories have built significant momentum — and where the policy debate over open-source AI remains unresolved in Washington.
Meta on Monday unveiled Muse Glimmer and announced it would open the weights of its latest model, Muse Spark 1.2. A day later, Nvidia debuted Nemotron 3.5 Lightning, which the chipmaker described as "truly open source" because it publishes the related training datasets, techniques, and model weights for developers to inspect. Both models are smaller than frontier offerings and are designed to run on laptops for tasks like powering on-device digital agents.
The releases come roughly three weeks after Meta and Nvidia were among more than 20 U.S. tech companies that jointly urged policymakers to avoid placing "premature restrictions" on open-weight models. "The age of AI can be one of prosperity," the consortium wrote in an open letter dated July 24. "With the right choices, open weight AI can expand opportunity, strengthen competition, extend American technological leadership, mitigate risk, and ensure that the benefits of this extraordinary technology are shared broadly across our economy."
The two companies are now competing with models from Chinese AI laboratories including Moonshot AI, DeepSeek, and Alibaba's Qwen — all of which are freely available to developers.
Box CEO Aaron Levie, one of the signatories of the July letter, described Zuckerberg's plan for Muse Spark 1.2 as "a very big deal," arguing it is a powerful model that rivals top foundation models from Anthropic and OpenAI. He said the domestic availability of such models opens meaningful opportunities in sectors where Chinese-developed alternatives are effectively off-limits. "You probably wouldn't be able to put a non-domestic open-source model in a major government agency, as an example, and you wouldn't be able to use it at very large banks most likely," Levie said. "There's a very firm flag in the ground that America will have near-frontier open-source models."
Meta's return to the open-weight market comes after a rocky stretch. The company's earlier open-source effort under the Llama branding drew developer criticism, particularly following the April 2025 release of Llama 4, which left many users unimpressed. Meta subsequently spent billions of dollars overhauling its AI division, installing Scale AI CEO Alexandr Wang as its leader. Wang's group has since been releasing proprietary models under the Muse brand in an effort to build new revenue streams.
That pivot away from open weights created friction with the developer community that Meta is now working to repair. Umesh Sachdev, CEO of business AI startup Uniphore, said the shift had damaged trust. "I think it's going to take more than a 3,500 worded article from Zuck to convince developers," Sachdev said, referring to an accompanying manifesto Zuckerberg published alongside this week's release. "The emotion of my developers at Uniphore, they almost feel betrayed."
Sachdev nonetheless expressed support for domestic competition. "More competition will drive down token cost, and will drive up innovation, and it's always good for consumers," he said.
Forrester analyst Charlie Dai offered a measured assessment, calling Meta's move "strategically important because it restores a major U.S. frontier AI vendor to the open ecosystem." Dai said developers and enterprises would likely welcome the shift back toward open weights for the transparency and deployment flexibility it affords, but added that Meta must now "prove it can cultivate a durable ecosystem beyond releasing competitive models."
Nemotron 3.5 Lightning stems from Nvidia's Nemotron 3 family of models, which the company first released in December. The chipmaker has framed its open approach as a differentiator from the proprietary model strategies pursued by OpenAI and Anthropic.
Whether the releases translate into meaningful developer adoption will depend partly on how convincingly Meta can rebuild goodwill with engineers who shifted to alternative platforms — and on how the policy environment in Washington evolves around Chinese open-weight models.
Sara Montes de Oca is the Editor in Chief of TechEchelon. Previously a correspondent and producer in Washington, D.C., covering business, finance, and politics.
More from Sara →Brad Lightcap, OpenAI's former chief operating officer and an eight-year veteran of the company, announced Tuesday he is departing to start a new venture, adding to a string of senior leadership exits at the AI lab.
London's King's Cross neighborhood, once one of the city's most troubled districts, has emerged as a global AI hub rivaling San Francisco and Beijing, drawing OpenAI, Meta, Anthropic, and hundreds of startups as office vacancy rates fall to 0.9%.
Airbnb CEO Brian Chesky said AI has cut feature development time by as much as 60% and increased features shipped by nearly 80%, as the company prepares to begin testing a natural-language AI search toggle for users.
Written for readers who already know the basics — markets, AI, and the policy decisions that shape both.