DeepSeek V3.2 Upgrades to 1M Context Window — Available via API | Model Release
DeepSeek upgraded its flagship model to DeepSeek V3.2 (Dec 2025), extending the context window to 1 million tokens and updating the knowledge cutoff to May 202…
Published on MyPrivateClaw
Mar 31, 2026, 6:50 AM UTC
Coverage date
Mar 10, 2026
Last updated
Apr 4, 2026, 5:45 AM UTC
News summary
DeepSeek upgraded both its deepseek chat and deepseek reasoner API endpoints to DeepSeek V3.2 on December 1, 2025. The deepseek chat endpoint corresponds to V3.2's non thinking mode (fast, direct responses), while deepseek reasoner corresponds to its thinking mode (extended chain of thought reasoning). The upgrade followed an experimental V3.2 Exp deployment in September 2025 and a V3.1 Terminus release in September that addressed language consistency issues including Chinese English mixing and abnormal characters. DeepSeek V3.2 also incorporates the hybrid reasoning architecture introduced in V3.1, where a single model supports both thinking and non thinking modes without requiring separate model weights. Earlier in the V3 lineage, DeepSeek V3.1 (August 2025) brought significant agent capability improvements: SWE bench Verified score of 66.0, SWE bench Multilingual of 54.5, and Termina…