Overview

DeepSeek-V3.1 merged the reasoning behaviour of DeepSeek-R1 into the V3 architecture as a hybrid model: the same weights can answer quickly or think step by step, depending on the chat template.

Highlights

  • 671B total, 37B active parameters
  • Hybrid thinking and non-thinking modes
  • Improved tool calling and agent tasks
  • MIT license

Typical uses

Self-hosted assistants, coding agents and reasoning workloads.

Things to consider

  • Needs about 404 GB of memory to run locally.
  • Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.

Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.