Nov 14, 05:42 PM

Nexusflow Launches Athene-V2, a 72B Model Competing with GPT-4o, Surpassing It in Function Calls by 18%

Nexusflow has launched Athene-V2, a new open model suite with 72 billion parameters, aimed at competing with OpenAI's GPT-4o across various benchmarks. Athene-V2 reportedly excels in chat, code, and math tasks, surpassing GPT-4o in single function calls by 18%. The model's performance in specific benchmarks includes scores of 84.9 for GPT-4o, 77.9 for Athene-V2, and 69.3 for Llama 3.1 405B in the Arena Hard benchmark, while in the Bigcode-Bench Hard benchmark, Athene-V2 scored 31.4 compared to GPT-4o's 30.8. Additionally, Athene-V2 is designed to address the current trends in AI development by focusing on targeted post-training to enhance model capabilities. The model is now available for use in LocalAI, where users can access its various functionalities, including the Athene-v2-agent and Athene-v2-chat models, which are specifically tailored for agentic and conversational tasks.

#Nexusflow #OpenAI #Llama #Arena Hard #AI #LocalAI #Athene

Written with ChatGPT (GPT-4o mini).

Sources

Additional media

Image #1 for story nexusflow-launches-athene-v2-72b-model-competing-gpt-4o-surpassing-function-18-c9ee4279

Image #2 for story nexusflow-launches-athene-v2-72b-model-competing-gpt-4o-surpassing-function-18-c9ee4279

Image #3 for story nexusflow-launches-athene-v2-72b-model-competing-gpt-4o-surpassing-function-18-c9ee4279

Image #4 for story nexusflow-launches-athene-v2-72b-model-competing-gpt-4o-surpassing-function-18-c9ee4279