Chinese AI startup DeepSeek has launched an upgraded R1 reasoning AI model, now featuring 685 billion parameters. This update enhances its performance to match significant competitors like OpenAI's o3 and Google's Gemini 2.5 Pro, with an accuracy increase from 70% to 87.5%. The advancements reduce the risk of hallucinations and improve the model's capabilities in generating code from natural language prompts, marking a notable improvement in AI resource efficiency and practical usage in programming environments.
At the core of DeepSeek's latest AI model, the R1 reasoning AI, is its dramatically increased parameter count, which has risen to a staggering 685 billion, effectively boosting its accuracy and performance.
DeepSeek's R1 model is not just about scale; it's also about better resource efficiency and reduced hallucinations, showcasing a promising direction for AI development and application.
Collection
[
|
...
]