DeepSeek V3.2 vs DeepSeek V4 Flash 0731

Change models
IndexAll categories, out of 100.
44.0 ±4.9
60.6 ±3.7 best
GapIndex points below the top.
−16.6
Top
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
36/s
93/s best
EloLMArena rating and rank.
1425
—
ReleasedMonth of release.
2025-12
2026-07
Input / 1MUS dollars per 1M input tokens.
$0.26
$0.044 best
Output / 1MUS dollars per 1M output tokens.
$0.38
$0.30 best
Batch output / 1MBatch price per 1M output tokens.
—
$0.144
1K chats2K tokens in, 500 out, each.
$0.71
$0.239 best
1K documents20K tokens in, 1K out, each.
$5.58
$1.19 best
1K agent tasks200K tokens in, 20K out, each.
$59.60
$14.86 best
ContextMaximum tokens in one request.
164K
1.05M best
AgenticMulti-step tasks with tools.
43.1
58.7 best
CodingCode writing and repair.
48.4
62.0 best
ReasoningLogic problems and puzzles.
42.7
64.3 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
43.3
58.1 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
35.0
58.5 best
MathMath problems.
46.4
58.7 best

Benchmarks

16 shared
Only DeepSeek V3.2: 22Only DeepSeek V4 Flash 0731: 69
DeepSeek V3.2
DeepSeek V4 Flash 0731
77.3%99 / 220
91.0% best37 / 220
11.2%132 / 216
38.6% best53 / 216
45.7%157 / 211
79.7% best45 / 211
0.9%119 / 207
16.6% best45 / 207
68.4%106 / 203
94.4% best50 / 203
75.1%112 / 193
90.8% best30 / 193
24.0%111 / 170
40.4% best53 / 170
66.0%135 / 149
77.7% best107 / 149
69.9%91 / 142
87.3% best16 / 142
4.0%91 / 137
61.4% best40 / 137
76.3%78 / 135
89.9% best27 / 135
MMLU ProKnowledge
83.1%74 / 135
86.2% best45 / 135
57.0%76 / 132
89.0% best42 / 132
CorpFin v2Finance
47.9%110 / 131
61.8% best51 / 131
3.9%80 / 127
63.9% best36 / 127
61.6%71 / 116
94.8% best38 / 116