Loading...
Loading...
Browse all stories on DeepNewz
VisitChina's StepFun Unveils Step-2, Trillion-Parameter MoE Model Developed in Two Months, Ranks Fifth Globally
Nov 21, 2024, 05:00 AM
Chinese artificial intelligence startup StepFun has developed Step-2, a trillion-parameter Mixture of Experts (MoE) language model with a 16k context length that ranks fifth globally on LiveBench evaluations. The model has surpassed GPT-4o and is just behind o1-mini, demonstrating China's rapid progress in AI technology despite GPU trade blockades. Remarkably, Step-2 was developed in just two months and trained at a fraction of the cost of OpenAI's GPT-4, with an estimated expenditure of $3 million compared to GPT-4's $80–$100 million. The emergence of Step-2, China's top-performing large language model, highlights the country's significant investments in AI and its growing competitiveness in the field.
View original story
Cerebras Systems • 25%
Meta • 25%
OpenAI • 25%
Other • 25%
Finance • 25%
Healthcare • 25%
E-commerce • 25%
Other • 25%
Milestone 1: Initial phase completion • 25%
Milestone 2: Mid-phase completion • 25%
Milestone 3: Final phase initiation • 25%
No milestones announced • 25%
Meta • 25%
OpenAI • 25%
Anthropic • 25%
Other • 25%
Google • 25%
Meta • 25%
Microsoft • 25%
Other • 25%
Claude 3.5 • 25%
GPT-4 • 25%
Gemini Pro • 25%
Llama • 25%
Yes • 50%
No • 50%
Yes • 50%
No • 50%
LTM-2-Mini Pro • 25%
LTM-3 • 25%
LTM-2-Mini Ultra • 25%
Other • 25%
StepFun • 25%
Other • 25%
o1-mini • 25%
OpenAI • 25%
Other • 25%
China • 25%
USA • 25%
European Union • 25%