DeepSeek V3.1 vs Gemma 4 12B

Change models
IndexAll categories, out of 100.
38.0 ±5.8
38.3 ±3.9 best
GapIndex points below the top.
−0.3 · no clear gap
Top
CoverageIndex weight that has results.
85%
100% best
SpeedOutput tokens per second.
98/s
—
EloLMArena rating and rank.
1420
—
ReleasedMonth of release.
2025-08
2026-06
Input / 1MUS dollars per 1M input tokens.
—
—
Output / 1MUS dollars per 1M output tokens.
—
—
1K chats2K tokens in, 500 out, each.
—
—
1K documents20K tokens in, 1K out, each.
—
—
1K agent tasks200K tokens in, 20K out, each.
—
—
ContextMaximum tokens in one request.
128K
256K best
AgenticMulti-step tasks with tools.
38.3 best
31.0
CodingCode writing and repair.
42.7
42.7
ReasoningLogic problems and puzzles.
38.0
41.1 best
MultimodalTasks with images and text.
—
35.3
KnowledgeFacts and expert knowledge.
41.7 best
39.6
MultilingualTasks in many languages.
—
29.8
InstructionTasks with strict prompt rules.
24.9
56.8 best
MathMath problems.
29.9
40.7 best

Benchmarks

9 shared
Only DeepSeek V3.1: 8Only Gemma 4 12B: 20
DeepSeek V3.1
Gemma 4 12B
6.7%159 / 216
15.7% best122 / 216
47.0%155 / 211
63.7% best132 / 211
0.0%145 / 207
0.0%145 / 207
73.5%118 / 193
75.3% best109 / 193
23.1% best116 / 170
15.7%153 / 170
74.9%82 / 144
78.8% best71 / 144
34.8%103 / 143
36.3% best102 / 143
37.8%109 / 140
73.5% best27 / 140
83.7% best25 / 73
77.2%42 / 73