Meta and Nvidia release open-weight AI models to rival Chinese labs

Meta and Nvidia released open-weight artificial intelligence models on August 10 and 11, 2026, respectively, intensifying competition with Chinese labs that dominate the open-source AI arena. The models, Muse Glimmer and Nemotron 3.5 Lightning, are available for developers to download at no cost.

The releases follow a July 24, 2026 open letter in which the companies and other tech leaders urged U.S. policymakers to avoid “premature restrictions” on open-weight models, including those from China. The letter defended the practice of distillation, which critics view as intellectual property theft, as “a widely used technique for model improvement, evaluation, and validation.”

Muse Glimmer and Nemotron 3.5 Lightning are small models designed to run on laptops, powering on-device digital agents. Meta CEO Mark Zuckerberg separately announced plans to release the weights for a more powerful model, Muse Spark 1.2, which Box CEO Aaron Levie called a “very big deal” because it rivals top foundation models from OpenAI and Anthropic. Nvidia, building on its Nemotron 3 family from December 2025, said Nemotron 3.5 Lightning is “truly open source,” with published training datasets, techniques and weights.

Meta’s open-weight push comes after a rocky year. Its Llama 4 model, released in April 2025, disappointed developers and triggered a multibillion-dollar overhaul of the AI unit under Scale AI CEO Alexandr Wang. The unit then shifted to proprietary Muse-branded models, a move that Uniphore CEO Umesh Sachdev said “burned bridges” with third-party developers who “almost feel betrayed.”

Some analysts see strategic gains. “There’s a very firm flag in the ground that America will have near-frontier open-source models,” Levie said. He noted that government agencies and large banks, wary of Chinese models, can now adopt Meta’s offerings, unlocking “a tremendous amount of potential.” Forrester analyst Charlie Dai said Meta’s shift “restores a major U.S. frontier AI vendor to the open ecosystem” and will improve transparency and data sovereignty. Sachdev added that more competition will drive down token costs and boost innovation.

Yet both companies must prove durable demand. Dai cautioned that Meta must “cultivate a durable ecosystem beyond releasing competitive models,” while Nvidia enters a field crowded with popular Chinese models from Moonshot AI, DeepSeek and Alibaba’s Qwen.