AI21 Labs 后训练 NVIDIA Nemotron 3.5 Super，成本仅 GPT-6.1 Sol 的 1/25
AI21 后训练了 NVIDIA 的 Nemotron 3.5 Super，效果超过 GPT-6.1 Sol 但成本只有 1/25，还被 Jensen Huang 拿上主题演讲了。
AI21 Labs 的前向部署研究团队对 NVIDIA 的 Nemotron 3.5 Super VL 进行了后训练。该项目在任务质量上超过 GPT-6.1 Sol，成本仅为后者的 1/25。这一成果在 Jensen Huang 于 GTC Berlin 2026 的主题演讲中被提及。Jensen Huang 的核心观点是：后训练的开源模型能让企业建立主权智能，同时保持前沿级质量。
Proud to have our work post-training @nvidia Nemotron 3.5 Super VL with @alice_dot_io recognized on stage in @JensenHuang’s keynote, kicking off @NVIDIAGTC Berlin 2026🔥
Our team of Forward-Deployed Researchers post-trained recently released Nemotron 3.5 Super, surpassing GPT-6.1 Sol on task quality, while costing 25x less.
Jensen's message was clear: Post-trained open models enable companies to establish sovereign intelligence, without sacrificing frontier level quality. If you're ready to start this journey, book time with one of our training experts: https://t.co/Dn8LztfSem
