Inference
EN

DeepSeek-R1-0528 vs DeepSeek V3.1 (Reasoning)

Change models
DeepSeek-R1-0528DeepSeek
DeepSeek V3.1 (Reasoning)DeepSeek
IndexAll categories, out of 100.
44.6 ±8.9 best
43.0 ±6.4
GapIndex points below the top.
Top
−1.6 · no clear gap
CoverageIndex weight that has results.
60%
85% best
SpeedOutput tokens per second.
20/s
—
EloLMArena rating and rank.
1428
—
ReleasedMonth of release.
2025-05
2025-08
Input / 1MUS dollars per 1M input tokens.
$0.50
—
Output / 1MUS dollars per 1M output tokens.
$2.15
—
1K chats2K tokens in, 500 out, each.
$2.08
—
1K documents20K tokens in, 1K out, each.
$12.15
—
1K agent tasks200K tokens in, 20K out, each.
$143.00
—
ContextMaximum tokens in one request.
164K best
128K
AgenticMulti-step tasks with tools.
—
27.6
CodingCode writing and repair.
46.9
51.4 best
ReasoningLogic problems and puzzles.
38.2
43.5 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
46.0
47.0 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
27.8
MathMath problems.
45.9
46.5 best

Benchmarks

6 shared
Only DeepSeek-R1-0528: 7Only DeepSeek V3.1 (Reasoning): 9
DeepSeek-R1-0528
DeepSeek V3.1 (Reasoning)
Graduate-Level Google-Proof Q&AKnowledge
81.0% best64 / 141
80.1%66 / 141
Humanity's Last ExamKnowledge
17.7% best68 / 77
15.9%70 / 77
Massive Multitask Language Understanding ProfessionalKnowledge
85.0% best20 / 75
84.8%23 / 75
American Invitational Mathematics Examination 2025Math
87.5%22 / 53
88.4% best20 / 53
American Invitational Mathematics Examination 2024Math
91.4%6 / 27
93.1% best3 / 27
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for CodeCoding
73.3%4 / 12
74.8% best2 / 12