DeepSeek: DeepSeek V3.2
deepseek/deepseek-v3.2
Context: 163,840In: $0.27/1MOut: $0.40/1MUpdated 2026-08-05
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
Intelligence
32.0
Coding
44.2
Agentic
18.3
Metric profile
Trend over time
Intelligence
32.0
Coding
44.2
Agentic
18.3
Only 1 snapshot so far (data refreshes daily) — the trend line appears after the 2nd snapshot.
Classic evals (curated from model cards)
DeepSeek — manually maintained, may lag releases
MMLU-Pro
75%
GPQA Diamond
67%
MATH-500
95%
Design Arena (ELO by category)
| Category | ELO | Win rate | Rank |
|---|---|---|---|
| website | 1198 | 50.2% | #57 |
| codecategories | 1193 | 49.3% | #58 |
| 3d | 1188 | 49.5% | #50 |
| dataviz | 1185 | 48.1% | #57 |
| uicomponent | 1184 | 46.8% | #56 |
| gamedev | 1176 | 46.6% | #59 |
| asciiart | 1120 | 40.5% | #50 |
| svg | 1079 | 40.8% | #59 |