We provide the latest news
from the world of economics and financeNvidia CEO Jensen Huang speaks at CES in January. DeepSeek may give Nvidia an opportunity to expand its budding consumer AI market. Photo: AFP/Getty Images
Chinese startup DeepSeek has achieved what many (including experts) thought impossible: create a foundational, competitive AI model for just $5.6 million. By comparison, OpenAI’s unreleased GPT-5 (codenamed “Orion”) reportedly requires $500 million for six months of training.
While OpenAI often relies on supervised fine-tuning and massive computational resources, DeepSeek has pioneered a more efficient approach through pure reinforcement learning (RL), centered around the startup’s innovative GRPO (Group Reinforcement Policy Optimization) algorithm.
Jurica Dujmovic is a columnist for MarketWatch. He is a business publisher, consultant, designer and gamer. Follow him on Twitter @JuricaDujmovic.
Intraday Data provided by FACTSET and subject to terms of use. Historical and current end-of-day data provided by FACTSET. All quotes are in local exchange time. Real-time last sale data for U.S. stock quotes reflect trades reported through Nasdaq only. Intraday data delayed at least 15 minutes or per exchange requirements.