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 that reduces training and inference cost while preserving quality in long-context scenarios.
Input
Output
Context
164K
Max Output
66K
Parameters
671B
Technical Specifications
Model TypeChat
Context Window164K
Max Output Tokens66K
Parameters671B
Training Cutoff—
Licensemit
Capabilities
Input Modalities
text
Output Modalities
text
Features
code_generationfunction_callingjson_modereasoningstreamingtext_generation
Benchmarks
22.2%
HLE
84.0%
GPQA
38.9%
SciCode
86.2%
MMLU Pro
92.0
Math
36.7
Coding
86.2%
LiveCodeBench
41.7
Intelligence
Resources & Links
Related Tools
Data sourced from official provider APIs and documentation
Last updated: May 5, 2026
Start building with the right model.
From model selection to production, one platform, no fragmentation.