Skip to main content

Model profile

Kimi K3

Moonshot AICurrentReleased Jul 16, 2026
Data verified
Overall Score
80.96Public #4 of 200Verified #4 of 99
Arena Elo
1487
Eligible category ranks
4of 8
Price (1M tokens)
$3 in / $15 out
API pricing
Speed
Not listed
Context
Not available

Evidence coverage

57 of 323 tracked benchmarks are published. 34 are verified and 23 provisional. 5 of 8 categories are measured.

Updated July 23, 2026Methodology
Published / tracked
57 / 323
Verified
34
Provisional
23
Categories with evidence
5 / 8

Evidence by category

  • Agentic21 benchmarks
    Mixed evidence
  • Coding9 benchmarks
    Mixed evidence
  • Reasoning2 benchmarks
    Reported
  • Knowledge10 benchmarks
    Mixed evidence
  • Math0 benchmarks
    Not measured
  • Multilingual0 benchmarks
    Not measured
  • Multimodal15 benchmarks
    Mixed evidence
  • Inst. Following0 benchmarks
    Not measured
PendingReasoning
Confidence:
Low
base

Kimi K3 ranks #4 out of 200 models on the public leaderboard with an overall score of 80.96/100. It also ranks #4 out of 99 on the verified leaderboard. This places it among the top tier of AI models available in 2026, competing directly with the strongest models from leading AI labs.

Kimi K3's weight availability, context window, and reasoning mode have not been published in a source we can verify yet.

BenchLM links it directly to Kimi K2.6 as the earlier related model in that lineage. This profile currently has 57 of 323 tracked benchmarks. BenchLM only exposes non-generated benchmark rows publicly, so missing categories stay blank until a sourced evaluation is available.

Its strongest category is Multimodal & Grounded (#2), while its weakest is Knowledge (#4). This performance profile makes it particularly strong for screenshots, documents, charts, and grounded multimodal workflows.

Peer position

Exact provisional scores and ranks for the closest listed peers. A score can appear before a model clears the evidence threshold for a rank, so equal scores can have different rank states.

Range 76.7283.93

  1. Claude Mythos 5
    Anthropic
    #183.93
    Claude Mythos 5 is #1 with a score of 83.93.
    Compare
  2. Claude Fable 5
    Anthropic
    #283.68
    Claude Fable 5 is #2 with a score of 83.68.
    Compare
  3. GPT-5.6 Sol
    OpenAI
    #381.96
    GPT-5.6 Sol is #3 with a score of 81.96.
    Compare
  4. Kimi K3Current model
    Moonshot AI
    #480.96
    Kimi K3 is #4 with a score of 80.96.
  5. Claude Opus 4.8
    Anthropic
    #578.34
    Claude Opus 4.8 is #5 with a score of 78.34.
    Compare
  6. Muse Spark 1.1
    Meta
    #677.44
    Muse Spark 1.1 is #6 with a score of 77.44.
    Compare
  7. Grok 4.5
    xAI
    #776.72
    Grok 4.5 is #7 with a score of 76.72.
    Compare

Category percentile

More

Relative position among models eligible for each sourced category. A higher percentile means a stronger position within that category's ranked cohort; 100 is highest.

  1. Multimodal96%
    Eligible cohort rank #2 of 29Category score 90.3
  2. Knowledge94%
    Eligible cohort rank #4 of 52Category score 89.5
  3. Coding98%
    Eligible cohort rank #4 of 122Category score 78.0
  4. Agentic97%
    Eligible cohort rank #4 of 119Category score 66.6

Category evidence

Scores and ranks appear only where this model has published benchmark evidence. Categories without displayable source records remain not measured.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #4 of 119Percentile 97thWeight 22%21 benchmarksMixed sources66.6
CodingRank #4 of 122Percentile 98thWeight 20%9 benchmarksMixed sources78.0
ReasoningWeight 17%2 benchmarksReportedScore pending
KnowledgeRank #4 of 52Percentile 94thWeight 12%10 benchmarksMixed sources89.5
MathWeight 5%0 benchmarksNot measuredNot measured
MultilingualWeight 7%0 benchmarksNot measuredNot measured
MultimodalRank #2 of 29Percentile 96thWeight 12%15 benchmarksMixed sources90.3
Inst. FollowingWeight 5%0 benchmarksNot measuredNot measured

Chatbot Arena performance

Scroll horizontally to inspect confidence intervals and vote counts.

Chatbot Arena Elo, confidence interval, and vote count by evaluation view
ViewEloConfidence intervalVotes
Text Overall1487±10.03,614
Coding1531±19.7970
Instruction Following1480±17.91,131
Creative Writing1472±24.6609
Multi-turn1502±25.2585
Hard Prompts1506±12.62,312
Hard Prompts (English)1508±20.6850
Longer Query1504±16.21,528

Benchmark Details

Rows below have a displayable published verification record. Each source link and provenance note remains in the page HTML while its category is closed. Source-unverified manual rows and generated rows stay hidden.

Agentic21 benchmarks
Terminal-Bench 2.0Provider exact
88.3%Weighted 38%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 88.3 on Terminal-Bench 2.1 with the KimiCode harness at max reasoning effort.
BrowseCompProvider exact
91.2%Weighted 28%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 91.2% on BrowseComp with context compaction triggered at 300K tokens. Its 1M-context run without context management scores 90.4%.
DeepSearchQAProvider exact
95.0%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 95.0 F1 on DeepSearchQA.
Toolathlon-VerifiedProvider exact
73.2%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 73.2 on Toolathlon-Verified.
MCP AtlasProvider exact
84.2%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 84.2 on the 500-task public MCP Atlas subset with a 100-turn limit and Gemini 3.1 Pro as judge.
AutomationBenchProvider exact
30.8%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 30.8 on the 600-task public AutomationBench subset using the official GitHub setup.
JobBenchProvider exact
52.9%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 52.9 on JobBench.
APEX-AgentsProvider exact
37.6%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 37.6 on APEX-Agents.
SpreadsheetBench 2Provider exact
34.8%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 34.8 on SpreadsheetBench 2 with the Claude Code harness.
DECK-BenchProvider exact

DECK-Bench (Internal)

73.5%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 73.5 on its internal DECK-Bench.
AA Agentic IndexReported

Artificial Analysis Agentic Index

50.1%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
GDPval-AAReported

GDPval-AA normalized

59.0%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
GDPval-AAReported
1679Display only
Source: Artificial Analysis: gdpval-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA BriefcaseReported

Artificial Analysis Briefcase

1543Display only
Source: Artificial Analysis: aa-briefcase leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA AutomationBenchReported

Artificial Analysis AutomationBench

52.7%Display only
Source: Artificial Analysis: automationbench-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA EnterpriseOps-GymReported

Artificial Analysis EnterpriseOps-Gym

45.3%Display only
Source: Artificial Analysis: enterprise-ops-gym-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA Harvey LABReported

Artificial Analysis Harvey LAB-AA

94.6%Display only
Source: Artificial Analysis: harvey-lab-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA Tau3 BankingReported

Artificial Analysis Tau3-Banking

33.4%Display only
Source: Artificial Analysis: tau3-banking leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
aaTerminalBench21Reported
85%Display only
Source: Artificial Analysis: terminalbench-v2-1 leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
APEX-Agents-AAReported
41.3%Display only
Source: Artificial Analysis: apex-agents-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA ITBenchReported

Artificial Analysis ITBench-AA

47.7%Display only
Source: Artificial Analysis: itbench-aa leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Coding9 benchmarks
deepSweProvider exact
67.5%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 67.5 on DeepSWE with the KimiCode harness. The footnote separately reports 67.3 with mini-SWE-agent on the official DeepSWE leaderboard.
FrontierSWEProvider exact
81.2%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports an 81.2 FrontierSWE dominance score with the KimiCode harness, recomputed from raw scores with the official evaluation script on July 16, 2026.
ProgramBenchProvider exact

ProgramBench: Can Language Models Rebuild Programs From Scratch?

77.8%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 77.8 on Program Bench with the KimiCode harness.
Kimi Code Bench v2Provider exact
72.9%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 72.9 on the internal Kimi Code Bench 2.0 after evaluating K3 with KimiCode and Claude Code at max effort.
sweMarathonProvider exact
42%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 42.0 on SWE Marathon with the Claude Code harness.
PostTrain BenchProvider exact
36.6%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 36.6 on PostTrain Bench using the official Harbor implementation with Claude Code at max effort, averaged over three runs.
MLS-Bench LiteProvider exact
48.3%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 48.3 on MLS Bench Lite with the KimiCode harness.
AA Coding IndexReported

Artificial Analysis Coding Index

76.2%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-SciCodeReported

Artificial Analysis SciCode

58.7%Display only
Source: Artificial Analysis: scicode leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Reasoning2 benchmarks
AA-LCRReported

Artificial Analysis Long Context Reasoning

74.7%Display only
Source: Artificial Analysis: artificial-analysis-long-context-reasoning leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
CritPtReported

Critical Physics Tasks

23.4%Display only
Source: Artificial Analysis: critpt leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Knowledge10 benchmarks
HLEProvider exact

Humanity's Last Exam

56%Weighted 45%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 56.0% on HLE-Full with tools.
GPQAProvider exact

Graduate-Level Google-Proof Q&A

93.5%Weighted 7%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 93.5% on GPQA-Diamond; BenchLM maps the exact result to both the weighted GPQA lane and its display key.
GPQA-DProvider exact

GPQA Diamond

93.5%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 93.5% on GPQA-Diamond.
HLE w/o toolsProvider exact

Humanity's Last Exam without tools

43.5%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 43.5% on HLE-Full without tools.
Artificial Analysis Intelligence IndexReported
57.1%Display only
Source: Artificial Analysis: artificial-analysis-intelligence-index leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-GPQA DiamondReported

Artificial Analysis GPQA Diamond

93.5%Display only
Source: Artificial Analysis: gpqa-diamond leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-HLEReported

Artificial Analysis Humanity's Last Exam

44.3%Display only
Source: Artificial Analysis: humanitys-last-exam leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-Omniscience IndexReported

Artificial Analysis Omniscience Index

18.4%Display only
Source: Artificial Analysis: omniscience leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
AA-Omniscience AccuracyReported

Artificial Analysis Omniscience Accuracy

46.0%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
AA-Omniscience Hallucination RateReported

Artificial Analysis Omniscience Hallucination Rate

50.9%Display only
Source: Artificial Analysis model benchmarksProvenance: Display-only row synced from the current Artificial Analysis model payload. It is excluded from BenchLM weighted scoring.
Multimodal15 benchmarks
MMMU-ProProvider exact

Massive Multi-discipline Multimodal Understanding Pro

81.6%Weighted 45%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 81.6 on MMMU-Pro using the official protocol, preserving input order and prepending images to text.
OfficeQA ProProvider exact
63.3%Weighted 30%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 63.3 on OfficeQA Pro with the Claude Code harness.
CharXivProvider exact

CharXiv Reasoning

91.3%Weighted 25%
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 91.3 on CharXiv Reasoning with Python.
MMMU-Pro w/ PythonProvider exact

MMMU-Pro with Python

83.4%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 83.4 on MMMU-Pro with Python.
CharXiv w/o toolsProvider exact

CharXiv Reasoning without tools

84.8%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 84.8 on CharXiv Reasoning without Python.
MathVisionProvider exact
94.3%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 94.3 on MathVision without Python.
MathVision w/ PythonProvider exact

MathVision with Python

97.8%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 97.8 on MathVision with Python.
BabyVision w/ PythonProvider exact

BabyVision with Python

85.7%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 85.7 on BabyVision with Python.
ZeroBenchProvider exact
23.0%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 23.0 pass@5 on ZeroBench_main using the official setting over five runs.
ZeroBench w/ PythonProvider exact

ZeroBench_main with Python

41.0%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 41.0 pass@5 on ZeroBench_main with Python using the official setting over five runs.
WorldVQA ForceAnswerProvider exact
51.0%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 51.0 on WorldVQA ForceAnswer.
OmniDocBenchProvider exact
91.1%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 91.1 on OmniDocBench; BenchLM keeps it separate from the versioned OmniDocBench 1.5 key.
PerceptionBenchProvider exact

PerceptionBench (Internal)

58.5%Display only
Source: Moonshot AI: Kimi K3 launch blogProvenance: Moonshot reports 58.5 on its internal PerceptionBench.
AA-MMMU-ProReported

Artificial Analysis MMMU-Pro

80.5%Display only
Source: Artificial Analysis: mmmu-pro leaderboardProvenance: Display-only row synced from the current Artificial Analysis evaluation leaderboard. It is excluded from BenchLM weighted scoring.
Design Arena WebsiteReported

Design Arena Website Elo

1386Display only
Source: OpenRouter model benchmarksProvenance: Display-only Design Arena Website Elo synced from OpenRouter model benchmark metadata. It is excluded from BenchLM weighted scoring.

Kimi K3 Family

Base entry

Related Earlier Model

Kimi K2.6

Frequently Asked Questions

How does Kimi K3 perform overall in AI benchmarks?

Kimi K3 currently ranks #4 out of 200 models on BenchLM's provisional leaderboard with an overall score of 80.96 (estimated). It also ranks #4 out of 99 on the verified leaderboard. It is created by Moonshot AI.

Is Kimi K3 good for knowledge and understanding?

Kimi K3 ranks #4 out of 52 models in knowledge and understanding benchmarks with an average score of 89.5. It is among the top performers in this category.

Is Kimi K3 good for coding and programming?

Kimi K3 ranks #4 out of 122 models in coding and programming benchmarks with an average score of 78. It is among the top performers in this category.

Is Kimi K3 good for reasoning and logic?

Kimi K3 has visible benchmark coverage in reasoning and logic, but BenchLM does not currently assign it a global category rank there.

Is Kimi K3 good for agentic tool use and computer tasks?

Kimi K3 ranks #4 out of 119 models in agentic tool use and computer tasks benchmarks with an average score of 66.6. It is among the top performers in this category.

Is Kimi K3 good for multimodal and grounded tasks?

Kimi K3 ranks #2 out of 29 models in multimodal and grounded tasks benchmarks with an average score of 90.3. It is among the top performers in this category.

Does Kimi K3 have full benchmark coverage on BenchLM?

Not yet. Kimi K3 currently has 57 published benchmark scores out of the 323 benchmarks BenchLM tracks. BenchLM only exposes non-generated public benchmark rows, so missing categories stay blank until a sourced evaluation is available.

What is the context window size of Kimi K3?

Kimi K3's context window has not been published in a source we can verify yet. The profile leaves this field unavailable instead of borrowing a limit from an earlier model in the family.

Last updated: July 23, 2026 · Runtime metrics stay blank until BenchLM has a sourced snapshot.

Choose with this week’s evidence

Join 2,000+ readers for ranking moves, new releases, pricing changes, and the evidence behind them.

Free. One email per week.