ModelsCompareBest forBenchmarksStatusPricingAPI

Kimi K3

Moonshot AI·Operational·Released 2026-07-16textvision
Context1,000k
Max output1,048.576k
Input / 1M$3 ↗
Output / 1M$15 ↗
Cached in$0.3
Benchmarks59

Benchmarks

Cited public sources · rank is exact-benchmark, apples-to-apples

AIME 2026 (MathArena)#6/2196.7% ↗
FrontierMath Tier 4 v2 (Epoch AI run)#14/4439% ↗
FrontierMath Tiers 1-3 v2 (Epoch AI run)#14/5972.2% ↗
GPQA Diamond (Epoch AI run)#13/7793.1% ↗
OTIS Mock AIME 2024-2025 (Epoch AI run)#14/7697.2% ↗
SimpleQA Verified (Epoch AI run)#14/5350.6% ↗
AA-Briefcase#2/21548 ↗
AA-LCR#1/474.7% ↗
Agents' Last Exam#2/228.3% ↗
APEX-Agents#2/341% ↗
AutomationBench#2/230.8% ↗
BabyVision (with Python)85.7% ↗
BrowseComp (1M context, no compaction)90.4% ↗
BrowseComp (context compaction)91.2% ↗
CharXiv Reasoning (no tools)#3/384.8% ↗
CharXiv Reasoning (with Python)91.3% ↗
CorpFin v271.6% ↗
CritPt23.4% ↗
DeepSearchQA (F1)95% ↗
DeepSWE v1.1 (Kimi Code)67.5% ↗
Finance Agent v254.4% ↗
FrontierSWE#1/381.2% ↗
GDPval-AA v2#5/71686 ↗
GPQA Diamond#8/7093.5% ↗
Harvey Lab-AA94.6% ↗
Humanity's Last Exam (no tools)#5/2943.5% ↗
Humanity's Last Exam (with tools)#6/1356% ↗
JobBench#2/254.3% ↗
Kimi Code Bench 2.072.9% ↗
Legal Research Bench44.2% ↗
MathVision (no tools)94.3% ↗
MathVision (with Python)97.8% ↗
MCP-Atlas#1/384.2% ↗
MCPMark-Verified#1/394.5% ↗
MLS-Bench-Lite#1/348.3% ↗
MMMU-Pro (no tools)#1/681.6% ↗
MMMU-Pro (with Python)83.4% ↗
MMVU#1/282.1% ↗
OfficeQA Pro#1/263.3% ↗
OmniDocBench91.1% ↗
OSWorld 2.058.3% ↗
OSWorld-Verified#1/1284.8% ↗
PerceptionBench58.5% ↗
PostTrainBench#2/336.6% ↗
ProgramBench#1/377.8% ↗
ResearchRubrics#1/376.2% ↗
SaaS-Bench60.1% ↗
SciCode#1/558.7% ↗
SpreadsheetBench 234.8% ↗
SWE-Marathon42% ↗
tau3-Banking33.4% ↗
Terminal-Bench 2.1#2/1788.3% ↗
Toolathlon-Verified76.5% ↗
Video-MME (with subtitles)90% ↗
WorldVQA ForceAnswer51% ↗
ZeroBench pass@5 (no tools)23% ↗
ZeroBench pass@5 (with Python)41% ↗
BrowseComp (1M context, no compaction)90.4% ↗
DeepSWE v1.1#5/1367.3% ↗

Compare

vs GPT-6 Astravs GPT-5.6 Solvs GPT-5.6 Terravs GPT-5.6 Lunavs Claude Fable 5.1vs Claude Opus 5vs Claude Sonnet 5vs Claude Haiku 4.5

More from Moonshot AI

Kimi K2.7 CodeKimi K2.6

Access Kimi K3 and every other model through one endpoint with automatic failover: Respan gateway.