Prime Intellect separately worked with NVIDIA to post-train Nemotron 3.5 Super VL on thousands of tasks across finance, sales, operations, and marketing, improving its AutomationBench-AA score by about 70%. AI21 Labs stated that these results demonstrate the ability of companies to achieve AI independence through sovereign specialized intelligence, reducing their reliance on broad, general-purpose models.
Tuesday, October 20, 2026
Story
AI21 Beats Claude Opus 5.5 at 33x Lower Cost with Nemotron 3.5
AI5 days ago
AI21 Labs spent three weeks collaborating with firms like Gong and Deel to post-train NVIDIA's latest Super VL model into specialized tools for corporate use. The resulting models reached performance levels that meet or exceed the quality of general frontier AI, but at a smaller cost. A specialized model for Deel achieved 88.5% accuracy on an 8-country payroll benchmark, performing 8% better than Claude Opus 5.5 while running at 33x lower cost, while a cybersecurity model for Alice improved attack resilience by 41 points.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.