AI BriefingDeepSeek-V3.2 Released
Key point
DeepSeek unveiled the reasoning-focused models V3.2 and Speciale.
Details
DeepSeek-V3.2 has been released as the official successor to V3.2-Exp, and is available immediately on App, Web, and API.
Also announced alongside it is V3.2-Speciale, a version that maximizes reasoning performance, currently API-only. It's introduced as having achieved gold-medal-level results at IMO, CMO, ICPC World Finals, IOI 2025, but is described as a model that excels at complex problems in exchange for using more tokens.
The core change is Thinking in Tool-Use. DeepSeek introduced a large-scale agent training data synthesis method involving 1,800+ environments and 85k+ complex instructions, and is positioning V3.2 as the first model to directly integrate the thinking process into tool-use.
On the API side, V3.2 is provided in the same way as V3.2-Exp. V3.2-Speciale is provided via a temporary endpoint, available until December 15, 2025, 15:59 UTC, with pricing the same as V3.2, but tool calls are not supported. Additionally, V3.2 supports Thinking in Tool-Use.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis PerformanceDeepSeek Releases DSec: Elastic Sandbox Infrastructure for Large-Scale Agentic Training