DeepSeek V3.1 vs DeepSeek V3.2 (Thinking)
Change modelsDeepSeek V3.1DeepSeek
DeepSeek V3.2 (Thinking)DeepSeek
IndexAll categories, out of 100.
37.9 ±5.8
49.4 ±8.6 best
GapIndex points below the top.
−11.5
Top
CoverageIndex weight that has results.
85% best
65%
SpeedOutput tokens per second.
98/s
—
EloLMArena rating and rank.
1420
1420
ReleasedMonth of release.
2025-08
2025-12
Input / 1MUS dollars per 1M input tokens.
—
$0.55
Output / 1MUS dollars per 1M output tokens.
—
$2.19
1K chats2K tokens in, 500 out, each.
—
$2.20
1K documents20K tokens in, 1K out, each.
—
$13.19
1K agent tasks200K tokens in, 20K out, each.
—
$153.80
ContextMaximum tokens in one request.
128K
128K
AgenticMulti-step tasks with tools.
38.0
47.7 best
CodingCode writing and repair.
43.0
49.2 best
ReasoningLogic problems and puzzles.
38.2
—
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
41.5
51.2 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
24.3
—
MathMath problems.
29.9
51.0 best
Benchmarks
5 sharedOnly DeepSeek V3.1: 12Only DeepSeek V3.2 (Thinking): 12
DeepSeek V3.1
DeepSeek V3.2 (Thinking)