DeepSeek-V3.2 Release
10 days ago
- #DeepSeek
- #AI
- #Reasoning
- DeepSeek-V3.2 and DeepSeek-V3.2-Speciale have been released, focusing on reasoning-first models for agents.
- DeepSeek-V3.2 is the official successor to V3.2-Exp, available on App, Web, and API.
- DeepSeek-V3.2-Speciale offers advanced reasoning capabilities, currently API-only.
- V3.2 provides balanced inference and length, comparable to GPT-5 performance.
- V3.2-Speciale excels in complex reasoning, rivaling Gemini-3.0-Pro, with gold-level results in competitions like IMO and ICPC.
- V3.2-Speciale requires higher token usage and is API-only for community evaluation.
- DeepSeek-V3.2 introduces a new agent training method with 1,800+ environments and 85k+ complex instructions.
- V3.2 integrates thinking into tool-use, supporting both thinking and non-thinking modes.
- API updates include V3.2-Speciale's temporary endpoint, available until Dec 15, 2025, with the same pricing as V3.2.
- Open-source releases include DeepSeek-V3.2 and V3.2-Speciale models and a tech report.