DeepSeek

DeepSeek-R1-0528

#2MMLU-Redux
#2MMLU-Pro
#3BrowseComp-zh

by DeepSeek

+
+
+
+
About

DeepSeek-R1-0528 represents a specific release iteration of the DeepSeek-R1 model, developed to incorporate refinements and improvements from ongoing training. Built to provide enhanced reasoning capabilities based on accumulated insights, it continues the evolution of DeepSeek's reasoning-focused architecture.

+
+
+
+
Pricing Range
Input (per 1M)$0.50 -$0.70
Output (per 1M)$2.15 -$2.50
Providers3
+
+
+
+
Timeline
AnnouncedMay 28, 2025
ReleasedMay 28, 2025
+
+
+
+
License & Family
License
MIT
Base ModelDeepSeek-R1
Performance Overview
Performance metrics and category breakdown

Overall Performance

16 benchmarks
Average Score
60.1%
Best Score
93.4%
High Performers (80%+)
6

Performance Metrics

Max Context Window
262.1K
Avg Throughput
30.7 tok/s
Avg Latency
1ms
+
+
+
+
All Benchmark Results for DeepSeek-R1-0528
Complete list of benchmark scores with detailed information
MMLU-Redux
text
0.93
93.4%
Self-reported
SimpleQA
text
0.92
92.3%
Self-reported
AIME 2024
text
0.91
91.4%
Self-reported
AIME 2025
text
0.88
87.5%
Self-reported
MMLU-Pro
text
0.85
85.0%
Self-reported
GPQA
text
0.81
81.0%
Self-reported
HMMT 2025
text
0.79
79.4%
Self-reported
LiveCodeBench
text
0.73
73.3%
Self-reported
Aider-Polyglot
text
0.72
71.6%
Self-reported
CodeForces
text
0.64
64.3%
Self-reported
Showing 1 to 10 of 16 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+