Jul 19, 06:18 PM

Open-Weight Chat Models Achieve Rank 1 and 77.8% Score in Performance Milestones

Recent advancements in open-weight chat models have been announced, showcasing significant improvements in performance and capabilities. DeepSeek-V2-Chat-0628 has achieved Rank 1 in the Open Weight Model category in Chatbot Arena and is ranked #11 overall, outperforming other open-source models. It also holds the #3 position in both the Coding Arena and Hard Prompts Arena. Additionally, Athene-Llama3-70B, a fine-tuned version of Llama-3-70B from Meta AI, has been released by NexusflowX. This model has set a new record on Arena-Hard with a score of 77.8%, approaching the performance of top proprietary models like Claude-3.5 and GPT-4o. ELO ratings for these models are expected soon. These developments highlight the potential of post-training in enhancing the capabilities of open-source models, making them competitive with leading proprietary solutions. API Access and Model Card for DeepSeek-V2-Chat-0628 are also available.

#Chat #Open Weight Model #Chatbot Arena #Coding Arena #Hard Prompts Arena #Llama #Meta AI #NexusflowX #Claude #API Access #Model Card

Written with ChatGPT (GPT-4o).

Sources

Additional media

Image #1 for story open-weight-chat-models-achieve-rank-1-77-8-score-performance-milestones

Image #2 for story open-weight-chat-models-achieve-rank-1-77-8-score-performance-milestones

Image #3 for story open-weight-chat-models-achieve-rank-1-77-8-score-performance-milestones

Open-Weight Chat Models Achieve Rank 1 and 77.8% Score in Performance Milestones

Sources

Additional media

Similar Stories