DeepSeek V3.1 vs DeepSeek V3.2

Change models
IndexAll categories, out of 100.
38.0 ±5.8
44.0 ±4.9 best
GapIndex points below the top.
−6.0 · no clear gap
Top
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
98/s best
36/s
EloLMArena rating and rank.
1420
1425 best
ReleasedMonth of release.
2025-08
2025-12
Input / 1MUS dollars per 1M input tokens.
—
$0.26
Output / 1MUS dollars per 1M output tokens.
—
$0.38
1K chats2K tokens in, 500 out, each.
—
$0.71
1K documents20K tokens in, 1K out, each.
—
$5.58
1K agent tasks200K tokens in, 20K out, each.
—
$59.60
ContextMaximum tokens in one request.
128K
164K best
AgenticMulti-step tasks with tools.
38.3
43.1 best
CodingCode writing and repair.
42.7
48.4 best
ReasoningLogic problems and puzzles.
38.0
42.7 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
41.7
43.3 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
24.9
35.0 best
MathMath problems.
29.9
46.4 best

Benchmarks

10 shared
Only DeepSeek V3.1: 7Only DeepSeek V3.2: 28
DeepSeek V3.1
DeepSeek V3.2
6.7%159 / 216
11.2% best132 / 216
47.0% best155 / 211
45.7%157 / 211
0.0%145 / 207
0.9% best119 / 207
73.5%118 / 193
75.1% best112 / 193
23.1%116 / 170
24.0% best111 / 170
34.8%103 / 143
78.9% best68 / 143
37.8%109 / 140
49.0% best76 / 140
CorpFin v2Finance
51.5% best103 / 131
47.9%110 / 131
22.5%56 / 66
34.8% best42 / 66
41.3%18 / 47
50.0% best10 / 47