DeepSeekLanguage modelSupersededOpen weights
DeepSeek R1
The open reasoning model that shook markets in January 2025.
Overall Intelligence
#56
of 56
31.1 pts behind #1
0
races led
177
weeks at #1
57.6
score
- Released
- Jan 20, 2025
- 1.7 yr ago
- Context
- 128K
- 32K max output
- Price
- $0.55 / $2.19
- per 1M in/out · $0.96 blended
- Speed
- 30 tok/s
- 15s to first token
- Modalities
- text
- outputs text
- Weights
- Open
- 671B MoE (37B active)
The 30-second version
The January 2025 open reasoning model that shocked the industry with frontier-level results at a fraction of the training cost.
Key takeaways
- 1Open weights with a permissive licence.
- 2Triggered a market-wide rethink of AI economics.
- 3Showed reinforcement learning could teach reasoning efficiently.
Where it ranks
Current snapshot · illustrative scores · click any race to watch it
R1Best on each axis
Overall
Coding
54Coding+30.6 pts54Frontend & UI+190 Elo53Backend & APIs+27.3 pts54Distributed Systems+31.0 pts54Full-Stack Apps+29.9 pts52Debugging & Repair+28.5 pts54Agentic Coding+33.7 pts53Code Review & Security+27.2 pts53Data Science & SQL+27.4 pts54Mobile Apps+27.7 pts53Terminal & DevOps+36.8 pts51Competitive Programming+783
Hard Reasoning
Math
Science
Writing & Language
Multimodal
Tool Use & Agents
Speed & Cost
Overall Intelligence over time
#1 frontierR1V4.1 FlashGLM-5.3
Update history
The distilled version first; open any release for the deeper breakdown.
The DeepSeek V family