A new large language model, GigaChat-3.5 Reasoning, has been released. This model features a 432B-A28B Mixture of Experts (MoE) architecture with Gated DeltaNet for efficient long-context processing. It was trained using domain experts and distilled into a single model, reportedly achieving performance close to DeepSeek V4 Flash Preview while using significantly fewer tokens for reasoning. AI
IMPACT This release offers a new option for users seeking efficient long-context processing and potentially competitive reasoning capabilities.
RANK_REASON New model release from a frontier lab. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →