
DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference efficiency in long-context scenarios while maintaining output quality. Users can control the reasoning behaviour with the reasoning enabled boolean. Learn more in our docs(opens in new tab)
The model was trained under conditions aligned with V3.1-Terminus to enable direct comparison. Benchmarking shows performance roughly on par with V3.1 across reasoning, coding, and agentic tool-use tasks, with minor tradeoffs and gains depending on the domain. This release focuses on validating architectural optimizations for extended context lengths rather than advancing raw task accuracy, making it primarily a research-oriented model for exploring efficient transformer designs.
Modalities
In / Out Price
$0.27 / $0.41per 1M
Context
164K
Released
Sep 29, 2025
Knowledge Cutoff
Jul 2025
DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference efficiency in long-context scenarios while maintaining output quality.
DeepSeek V3.2 Exp costs $0.27/M input tokens and $0.41/M output tokens.
DeepSeek V3.2 Exp has a 163,840 token context window. It supports up to 65,536 completion tokens.
Yes. DeepSeek V3.2 Exp accepts tools and tool_choice for function calling. It also supports structured outputs via a JSON schema in response_format.
DeepSeek V3.2 Exp is served by 3 providers on OpenRouter: NovitaAI, SiliconFlow and AtlasCloud. Requests are routed to the best available provider, with automatic failover to the others, and you can pin or exclude providers with provider routing.
DeepSeek V3.2 Exp was released on September 29, 2025. Its knowledge cutoff is July 31, 2025.