Mixture of Experts GPT-5.2 code red & AWS Nova models drop
15 snips
Dec 12, 2025 
In this insightful discussion, Kate Soule, Director of Technical Product Management at Granite, emphasizes the importance of model transparency, sharing their impressive 95/100 score on the Stanford index. Ambhi Ganesan, an AI and analytics partner, analyzes enterprise adoption patterns and the impact of AWS Nova models, stressing strategic migration practices. Mihai Criveti, a distinguished engineer, critiques incremental updates in AI and advocates for the potential of long-running agents to execute complex tasks. Together, they unravel the balance between competition and consumer benefits in AI innovation.
AI Snips
Chapters
Transcript
Episode notes
Model Drops Often Maintain, Not Revolutionize
- Frequent model releases mostly tweak performance, speed, or cost rather than deliver revolutionary changes.
- Ambhi and Mihai argue these updates sustain competition but rarely transform consumer experiences immediately.
Benchmarks Misalign With Real Costs
- Benchmarks drive competitive releases but often misalign with real-world metrics like cost and energy efficiency.
- Stanford found local workloads can match performance at much lower energy and cost than large hosted models.
Switch Models Only For Step Changes
- Enterprises should avoid switching models for every new release and prioritize stability in production.
- Move only when you see a clear step-function improvement and have a maintenance roadmap.