DeepSeekLanguage modelSupersededOpen weights
DeepSeek V3.2
Sparse attention made long context nearly free; IMO-gold-level math variant.
Overall Intelligence
#41
of 56
19.1 pts behind #1
0
races led
45
weeks at #1
69.7
score
- Released
- Dec 1, 2025
- 11 mo ago
- Context
- 128K
- 64K max output
- Price
- $0.28 / $0.42
- per 1M in/out · $0.32 blended
- Speed
- 40 tok/s
- 3s to first token
- Modalities
- text
- outputs text
- Weights
- Open
- 671B MoE (37B active)
The 30-second version
Sparse attention made long context nearly free; IMO-gold-level math variant.
Key takeaways
- 1Sparse attention made long context nearly free; IMO-gold-level math variant.
Where it ranks
Current snapshot · illustrative scores · click any race to watch it
V3.2Best on each axis
Overall
Coding
39Coding+19.0 pts44Frontend & UI+132 Elo40Backend & APIs+16.9 pts38Distributed Systems+19.2 pts40Full-Stack Apps+18.7 pts39Debugging & Repair+18.8 pts38Agentic Coding+20.3 pts38Code Review & Security+16.9 pts40Data Science & SQL+17.3 pts37Mobile Apps+15.3 pts41Terminal & DevOps+24.2 pts36Competitive Programming+477
Hard Reasoning
Math
Science
Writing & Language
Multimodal
Tool Use & Agents
Overall Intelligence over time
#1 frontierV3.2V4.1 FlashGLM-5.3
Update history
The distilled version first; open any release for the deeper breakdown.
The DeepSeek V family