DeepSeek: DeepSeek V3.2
- by deepseek
- deepseek/deepseek-v3.2
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
What the numbers mean
- Context window163,840 tokens
- How much the model can read in one request, counting your prompt, any files and the conversation so far. That is about 120,000 words, or 246 pages.
- Max output147,456 tokens
- The longest single reply it can write: about 110,000 words.
- Input price$0.26 per million tokens
- What you pay for everything the model reads. A 1,000-word prompt costs less than a cent.
- Output price$0.38 per million tokens
- What you pay for everything the model writes. Output usually costs more than input. A 500-word answer costs less than a cent.
- Modalitiestext in, text out
- The kinds of content it accepts and the kinds it produces. A model that takes images in can read a screenshot; one that only puts text out cannot draw.
Supported parameters
The settings you can pass with a request. A model that lists tools can call functions; one that lists structured_outputs can be held to a JSON schema.
frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningrepetition_penalty