Claude Haiku 4.5 Thinking vs Kimi K2 Instruct

Change models
Kimi K2 InstructMoonshot AI
IndexAll categories, out of 100.
43.2 ±7.0 best
42.9 ±8.9
GapIndex points below the top.
Top
−0.3 · no clear gap
CoverageIndex weight that has results.
75% best
65%
SpeedOutput tokens per second.
—
—
EloLMArena rating and rank.
—
—
ReleasedMonth of release.
2025-10
—
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.
200K
—
AgenticMulti-step tasks with tools.
49.7 best
44.4
CodingCode writing and repair.
42.3 best
41.0
ReasoningLogic problems and puzzles.
—
—
MultimodalTasks with images and text.
30.4
—
KnowledgeFacts and expert knowledge.
43.3
43.3
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
51.0 best
42.7

Benchmarks

12 shared
Only Claude Haiku 4.5 Thinking: 16Only Kimi K2 Instruct: 3
Claude Haiku 4.5 Thinking
Kimi K2 Instruct
81.2%82 / 149
81.5% best81 / 149
TaxEval v2Finance
67.5%106 / 143
70.2% best93 / 143
41.2%129 / 142
70.4% best87 / 142
72.2% best88 / 135
71.5%91 / 135
MMLU ProKnowledge
78.7%100 / 135
79.4% best95 / 135
CorpFin v2Finance
60.6% best66 / 131
50.4%108 / 131
AIMEMath
82.7% best44 / 95
62.7%56 / 95
MedQAHealthcare
79.6%74 / 95
84.0% best62 / 95
MGSMMultilingual
92.1% best21 / 74
90.9%31 / 74
38.2% best36 / 66
25.8%51 / 66
IOI v1Coding
6.2% best43 / 61
1.3%58 / 61
50.0% best10 / 47
40.0%22 / 47