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 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.
Todos os provedores = compatível em todos os upstreams que servem este modelo. Alguns provedores = depende do upstream que atende a solicitação. Padrão = o valor enviado quando você não define nada.
| Parâmetro | Provedores | Padrão |
|---|---|---|
| frequency_penalty | Todos os provedores | Não enviado por padrão |
| include_reasoning | Todos os provedores | - |
| logit_bias | Alguns provedores | - |
| logprobs | Alguns provedores | - |
| max_tokens | Todos os provedores | - |
| min_p | Alguns provedores | - |
| presence_penalty | Alguns provedores | - |
| reasoning | Todos os provedores | - |
| repetition_penalty | Alguns provedores | - |
| response_format | Todos os provedores | - |
| seed | Alguns provedores | - |
| stop | Alguns provedores | - |
| structured_outputs | Todos os provedores | - |
| temperature | Todos os provedores | 0.6 |
| tool_choice | Todos os provedores | - |
| tools | Todos os provedores | - |
| top_k | Todos os provedores | - |
| top_logprobs | Alguns provedores | - |
| top_p | Todos os provedores | 0.95 |
