DeepSeek, R1 추론 모델 출시… OpenAI o1에 훨씬 저렴한 비용으로 도전장

DeepSeek-R1 fundamentally challenges the 'scaling hypothesis' that dominated AI investment in 2024.

Monday, February 2, 2026
2 min read
DeepSeek Official Blog
정규 소스
China
Full Analysis85%
LinkedInX
변경 사항

DeepSeek launched DeepSeek-R1, an open-source AI reasoning model, demonstrating competitive performance against OpenAI's o1 at a fraction of the training cost.

Key Figures
less than $6 millionDeepSeek-R1 reportedly cost less than $6 million to train.
37B active parametersThe model uses a mixture-of-experts architecture with only 37B active parameters.
671B totalThe model uses a mixture-of-experts architecture from 671B total parameters.
Source Report

중국 AI 연구소 DeepSeek이 오픈소스 추론 모델인 DeepSeek-R1을 공개했습니다. 이 모델은 여러 벤치마크에서 OpenAI의 o1과 동등하거나 그 이상의 성능을 보이며, 훈련 비용은 600만 달러 미만인 것으로 알려졌습니다. DeepSeek-R1은 새로운 MoE(Mixture-of-Experts) 아키텍처와 강화 학습 접근 방식을 활용하여 컴퓨팅 요구 사항을 획기적으로 줄였습니다. 이번 출시는 최첨단 AI 개발에 수십억 달러의 컴퓨팅 투자가 필요하다는 기존의 통념에 도전하며 AI 업계에 큰 파장을 일으키고 있습니다.

Sigvera Intelligence
1DeepSeek-R1 matches OpenAI o1 on MATH, AIME, and coding benchmarks at a fraction of the training cost
2The model uses a mixture-of-experts architecture with only 37B active parameters from 671B total
3Open-source release enables global researchers to build on frontier reasoning capabilities
Market Impact

DeepSeek-R1 fundamentally challenges the 'scaling hypothesis' that dominated AI investment in 2024. If frontier capabilities can be achieved at 1/100th the cost, the competitive landscape shifts dramatically — from capital-intensive to algorithm-intensive. This has immediate implications for APAC AI companies that lack access to massive GPU clusters.

AI & Frontier Intelligence

Where this signal fits in the broader landscape.

111 산업 시그널Product Launch
전체 보기
전체 보기
Verified from official source
PublisherDeepSeek Official Blog
게시일Feb 2, 2026
소스 유형Company Newsroom
소스 분류Verified Canonical
시그널 타임라인
최초 보도Feb 2, 2026
인덱싱Feb 3, 2026
게시Feb 4, 2026

https://api-docs.deepseek.com/news/news250120

Read Full Source
신뢰도:0.9%
Get cross-language signal intelligence

Get curated intelligence delivered to your inbox. No spam, unsubscribe anytime.

Sign in to save notes on signals.

로그인
기업DeepSeek산업AI & Frontier Intelligence지역China이벤트Product Launch출처공식

Stay ahead of the next signal.

Free weekly briefings with structured signal summaries. No spam, cancel anytime.