Inference
EN

Claude Sonnet 4 vs Kimi K2 Instruct

Change models
Claude Sonnet 4Anthropic
Kimi K2 InstructMoonshot AI
IndexAll categories, out of 100.
43.0 ±5.2
43.2 ±8.9 best
GapIndex points below the top.
−0.2 · no clear gap
Top
CoverageIndex weight that has results.
90% best
65%
SpeedOutput tokens per second.
49/s
—
EloLMArena rating and rank.
1339
—
ReleasedMonth of release.
2025-05
—
Input / 1MUS dollars per 1M input tokens.
$3.00
—
Output / 1MUS dollars per 1M output tokens.
$15.00
—
1K chats2K tokens in, 500 out, each.
$13.50
—
1K documents20K tokens in, 1K out, each.
$75.00
—
1K agent tasks200K tokens in, 20K out, each.
$900.00
—
ContextMaximum tokens in one request.
200K
—
AgenticMulti-step tasks with tools.
49.4 best
44.6
CodingCode writing and repair.
46.2 best
41.3
ReasoningLogic problems and puzzles.
38.9
—
MultimodalTasks with images and text.
49.4
—
KnowledgeFacts and expert knowledge.
44.5 best
43.6
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
—
—
MathMath problems.
38.9
43.3 best

Benchmarks

12 shared
Only Claude Sonnet 4: 18Only Kimi K2 Instruct: 3
Claude Sonnet 4
Kimi K2 Instruct
LegalBenchLegal
83.0% best62 / 149
81.5%81 / 149
TaxEval v2Finance
69.6%97 / 143
70.2% best93 / 143
LiveCodeBenchCoding
59.7%107 / 142
70.4% best87 / 142
GPQA Diamond · Vals AIKnowledge
69.4%95 / 135
71.5% best91 / 135
MMLU ProKnowledge
79.4%95 / 135
79.4%95 / 135
CorpFin v2Finance
54.7% best92 / 131
50.4%108 / 131
AIMEMath
38.5%71 / 95
62.7% best56 / 95
MedQAHealthcare
90.3% best46 / 95
84.0%62 / 95
MGSMMultilingual
93.0% best11 / 74
90.9%31 / 74
SWE-bench Verified · SWE-bench/experimentsCoding
74.8% best6 / 65
48.6%41 / 65
IOI v1Coding
6.5% best41 / 61
1.3%58 / 61
MATH 500Math
90.3%26 / 59
94.2% best10 / 59