0
Applied AI·August 12, 2026·1 min read

Nvidia is building a trillion-parameter open model, and it would still be smaller than China’s

Share

Nemotron 3.5 Lightning—a 30B MoE with only 3B active parameters and a 1M-token context—signals Nvidia is using open weights to anchor an ecosystem ahead of its planned trillion-parameter model. For builders, this is a viable backbone for long-context workloads today and a hint that model choice will increasingly be tied to your hardware vendor and toolchain, not just raw benchmark scores.

Applied AI

SpaceXAI debuts Grok 4.6, overtaking Kimi K3's performance and matching GPT-5.6 Sol for world's third best on Artificial Analysis

Grok 4.6 is optimized for long-running agents, coding, and knowledge work with a cheaper-to-run profile — that’s a direct bid for the high-CPU, always-on workflows enterprises are actually piloting now. If you’re building agentic systems, you now have another top-tier model to benchmark not just on quality but on sustained cost per task, not per token.