DeepSeek-v3.2-Speciale
10 days ago
- #AI
- #Machine Learning
- #DeepSeek-V3.2
- DeepSeek-V3.2 introduces three key technical breakthroughs: DeepSeek Sparse Attention (DSA), Scalable Reinforcement Learning Framework, and Large-Scale Agentic Task Synthesis Pipeline.
- DeepSeek-V3.2-Speciale surpasses GPT-5 and matches Gemini-3.0-Pro in reasoning proficiency.
- Achieved gold-medal performance in the 2025 International Mathematical Olympiad (IMO) and International Olympiad in Informatics (IOI).
- Released final submissions for IOI 2025, ICPC World Finals, IMO 2025, and CMO 2025 for community verification.
- Updated chat template includes 'thinking with tools' capability and a new 'developer' role for search agent scenarios.
- Local deployment recommendations include setting temperature = 1.0 and top_p = 0.95.
- DeepSeek-V3.2-Speciale is optimized for deep reasoning tasks and does not support tool-calling.
- Model weights are licensed under the MIT License.