SupraLabs has released SupraBrain-50M-v0.1, a hybrid language model that combines Gated DeltaNet linear recurrence with Sliding-Window Attention and Surprise-Gated update mechanisms. Despite being trained on significantly less data (5B tokens compared to 20B tokens), the model demonstrates performance comparable to Supra-Base-50M. The developers also announced that Supra2-100M is scheduled for release within hours. AI
IMPACT This release offers a new hybrid model architecture, potentially influencing future LLM development with its novel combination of recurrence and attention mechanisms.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
