Inference
EN

Kimi K3.1 Pro vs Kimi K3

Change models
Kimi K3.1 ProMoonshot AISmartest open
Kimi K3Moonshot AI
IndexAll categories, out of 100.
75.7 ±4.7 best
66.9 ±2.4
GapIndex points below the top.
Top
−8.8
CoverageIndex weight that has results.
80%
95% best
SpeedOutput tokens per second.
43/s
44/s best
EloLMArena rating and rank.
—
1474
ReleasedMonth of release.
2026-10
2026-07
Input / 1MUS dollars per 1M input tokens.
$2.00
$0.192 best
Output / 1MUS dollars per 1M output tokens.
$10.00 best
$16.00
Batch output / 1MBatch price per 1M output tokens.
$6.00 best
$11.40
1K chats2K tokens in, 500 out, each.
$9.00
$8.38 best
1K documents20K tokens in, 1K out, each.
$50.00
$19.84 best
1K agent tasks200K tokens in, 20K out, each.
$600.00
$358.40 best
ContextMaximum tokens in one request.
1.05M
1.05M
AgenticMulti-step tasks with tools.
76.7 best
65.9
CodingCode writing and repair.
71.9 best
66.2
ReasoningLogic problems and puzzles.
77.7 best
67.7
MultimodalTasks with images and text.
72.0 best
65.6
KnowledgeFacts and expert knowledge.
71.0 best
66.3
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
71.0
MathMath problems.
—
66.6

Benchmarks

33 shared
Only Kimi K3.1 Pro: 30Only Kimi K3: 96
Kimi K3.1 Pro
Kimi K3
Artificial Analysis Humanity's Last ExamKnowledge
52.5% best11 / 222
46.9%22 / 222
Artificial Analysis Long Context ReasoningReasoning
86.3%3 / 217
88.7% best1 / 217
Critical Physics TasksReasoning
24.6% best22 / 214
23.4%23 / 214
Artificial Analysis Omniscience AccuracyKnowledge
48.9% best23 / 164
47.6%28 / 164
Graduate-Level Google-Proof Q&AKnowledge
94.2% best5 / 141
93.5%10 / 141
GDPval-AA normalizedAgentic
67.3% best3 / 133
51.6%26 / 133
Artificial Analysis SciCodeCoding
60.8% best9 / 114
59.5%11 / 114
Artificial Analysis MMMU-ProMultimodal
83.9% best9 / 113
80.5%18 / 113
GPQA Diamond · BenchLMKnowledge
94.2% best6 / 84
93.5%10 / 84
Humanity's Last ExamKnowledge
70.0% best2 / 77
56.0%13 / 77
BrowseCompAgentic
92.1% best3 / 63
91.2%7 / 63
DeepSWEAgentic
75.5% best5 / 59
67.5%28 / 59
Humanity's Last Exam without toolsKnowledge
51.1% best8 / 53
43.5%19 / 53
Massive Multi-discipline Multimodal Understanding ProMultimodal
84.1% best2 / 53
81.6%7 / 53
CharXiv ReasoningMultimodal
93.4% best3 / 49
91.3%5 / 49
MazeBench: visual spatial reasoning in a 3D open worldAgentic
2.0% best15 / 48
1.0%16 / 48
FrontierCode 2.0Coding
19.6% best22 / 27
17.7%24 / 27
Artificial Analysis AnalystAgentAgentic
38.8%23 / 26
38.8%23 / 26
Bug Hunt BenchCoding
36.0% best11 / 25
21.0%20 / 25
Artificial Analysis DeepsecBench-AAAgentic
36.7% best3 / 21
12.5%17 / 21