DeepSeek V3.1 (Reasoning) vs Qwen3.5-4B

Change models
Qwen3.5-4BAlibaba
IndexAll categories, out of 100.
42.9 ±6.4
43.0 ±7.1 best
GapIndex points below the top.
−0.1 · no clear gap
Top
CoverageIndex weight that has results.
85% best
80%
SpeedOutput tokens per second.
—
33/s
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2025-08
—
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
262K best
AgenticMulti-step tasks with tools.
27.6
63.6 best
CodingCode writing and repair.
51.2
—
ReasoningLogic problems and puzzles.
43.3
43.5 best
MultimodalTasks with images and text.
—
35.5
KnowledgeFacts and expert knowledge.
46.9 best
41.0
MultilingualTasks in many languages.
—
35.5
InstructionTasks with strict prompt rules.
27.8
37.6 best
MathMath problems.
46.4 best
41.2

Benchmarks

6 shared
Only DeepSeek V3.1 (Reasoning): 9Only Qwen3.5-4B: 11
DeepSeek V3.1 (Reasoning)
Qwen3.5-4B
14.3% best128 / 222
9.9%145 / 222
56.7%144 / 217
63.0% best138 / 217
2.0% best105 / 214
0.0%150 / 214
77.9% best106 / 199
77.1%108 / 199
37.4%105 / 149
92.1% best35 / 149
41.5%98 / 146
52.0% best78 / 146