Inference
EN

Qwen3 235B-A22B vs QwQ-32B

Change models
Qwen3 235B-A22BAlibaba
QwQ-32BAlibaba
IndexAll categories, out of 100.
38.1 ±8.8 best
36.3 ±9.5
GapIndex points below the top.
Top
−1.8 · no clear gap
CoverageIndex weight that has results.
55% best
50%
SpeedOutput tokens per second.
44/s
—
EloLMArena rating and rank.
1366 best
1329
ReleasedMonth of release.
2025-04
2025-03
Input / 1MUS dollars per 1M input tokens.
$0.455
—
Output / 1MUS dollars per 1M output tokens.
$1.82
—
1K chats2K tokens in, 500 out, each.
$1.82
—
1K documents20K tokens in, 1K out, each.
$10.92
—
1K agent tasks200K tokens in, 20K out, each.
$127.40
—
ContextMaximum tokens in one request.
131K
131K
AgenticMulti-step tasks with tools.
—
—
CodingCode writing and repair.
44.0
—
ReasoningLogic problems and puzzles.
—
25.6
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
41.1 best
38.4
MultilingualTasks in many languages.
42.0 best
33.0
InstructionTasks with strict prompt rules.
30.7 best
28.7
MathMath problems.
32.2
41.9 best

Benchmarks

12 shared
Only Qwen3 235B-A22B: 12Only QwQ-32B: 5
Qwen3 235B-A22B
QwQ-32B
GPQA diamond · Epoch AIKnowledge
70.7% best113 / 217
65.3%124 / 217
Graduate-Level Google-Proof Q&AKnowledge
62.9%108 / 141
65.6% best102 / 141
Instruction-Following EvalInstruction
83.2%46 / 70
83.9% best42 / 70
American Invitational Mathematics Examination 2025Math
24.7%50 / 53
69.5% best37 / 53
American Invitational Mathematics Examination 2024Math
40.1%23 / 27
79.5% best15 / 27
MATH-500 Problem SetMath
91.2%20 / 27
98.0% best4 / 27
MMLU-ReduxKnowledge
89.2%12 / 19
90.0% best11 / 19
MMMLUKnowledge
79.8%13 / 15
80.9% best11 / 15
C-EvalKnowledge
86.1%10 / 12
88.4% best8 / 12
INCLUDEMultilingual
75.6% best9 / 12
69.7%12 / 12
Berkeley Function Calling Leaderboard v3Agentic
68.0%
66.4%
LiveCodeBench v5Coding
35.3%
62.7%