DeepSeek V3.2-Speciale

DeepSeek llm

Deep-reasoning variant of V3.2 that drops tool calling in favour of maximum reasoning depth, targeting maths and competitive programming.

overall ⌄

49.0

39 of 93 ranked

price · $/M tokens

self-hosted

open weights, no first-party API

to run it yourself ⌄

411 GB

minimum · 1.6 TB recommended

Benchmarks

What its maker published, and what anyone else measured. Every figure links the document it came from.

benchmark	published	measured
HLE	30.6	—

org	DeepSeek
released	2025-11-28
params	685B
license	mit
hf id	deepseek-ai/DeepSeek-V3.2-Speciale