DeepSeek ships V3.2 and a high-compute V3.2-Speciale for hard reasoning
DeepSeek released DeepSeek-V3.2 as its new production model and V3.2-Speciale as a high-compute variant aimed at hard reasoning problems. The release builds on DeepSeek Sparse Attention for efficient long contexts and adds reasoning combined with tool use for agent workloads. For practitioners, it continued DeepSeek's pattern of low-cost, openly released models competing with frontier systems.
Read the full story at DeepSeek ↗Like, save or discuss on Buzz
