Skip to main content
BenchLM
Data

Data as of November 17, 2026 · How the score is built

Superseded.Moonshot AI has newer models in this line:Kimi K3.1 Pro

Kimi K3

SupersededOpen WeightReasoning

Decision readingKimi K3 scores 67 out of 100 and ranks #20 of 218. This profile shows 77 source-displayable benchmark rows; its strongest eligible category is Multimodal & Grounded at #19. API pricing is $3 input and $15 output per million tokens, with cached input at $0.3.

Email me when its price, rank or availability changes.

Capability

67/100

field median 49.5#20 of 218 ranked models

Public

#20of 218

Verified #11 of 79

Price

$3input / $15 output

input median $0.60cached $0.30 · blended $9

Speed

38tok/s

field median 89 tok/sFirst token 56.25 s

Context

1.05Mtokens

field median 256,000Reported for this model; direct source link not stored

Decision snapshot

Where this model sits in the ranked field, its strongest published evidence, and what to validate before choosing it. The marker compares it with the current ranked catalog, not with an absolute quality threshold.

Strongest published evidence

Multimodal & Grounded ranks #19. Particularly strong for screenshots, documents, charts, and grounded multimodal workflows.

Validate before choosing

77 published rows leave some tracked benchmark slots empty. Coding is its lowest eligible category at #23.

Source-linked · 77 displayable benchmark rows

Is Kimi K3 right for your job?

Its place on the LLM Selector’s job rankings, the same order the job pages publish. Describe your own job to see where it stands for that.

Kimi K3 has a newer version, Kimi K3.1 Pro. Where Kimi K3.1 Pro ranks by benchmark score:

Describe the task and constraints. This selector recommends models; it cannot carry out the task or access your files and accounts.

We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy

Category score record

Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.

Category scores, ranks, weighting, benchmark coverage, and evidence status
CategoryScore
AgenticRank #21 of 140Percentile 86thWeight 22%19 benchmarksMixed sources
65.1
CodingRank #23 of 163Percentile 86thWeight 20%18 benchmarksVerified
56.3
ReasoningRank #21 of 161Percentile 88thWeight 17%2 benchmarksVerified
65.4
MultimodalRank #19 of 67Percentile 73rdWeight 12%14 benchmarksVerified
66.7
KnowledgeRank #20 of 181Percentile 89thWeight 12%6 benchmarksVerified
67.5
MultilingualWeight 7%0 benchmarksNot measured
Not measured
Inst. FollowingWeight 5%1 benchmarkVerified
Score pending
MathWeight 5%0 benchmarksNot measured
Not measured

77 of 922 tracked benchmark slots have displayable evidence · bars run 0–100

Coverage details

How much of this is verified

Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.

  1. Agentic16/19 verified
  2. Coding18/18 verified
  3. Reasoning2/2 verified
  4. Multimodal14/14 verified
  5. Knowledge6/6 verified
  6. MultilingualNot measured
  7. Inst. Following1/1 verified
  8. MathNot measured
Verified sourceProvisionalNot measured

Capability shape

Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.

Kimi K3 category percentile values

  • Agentic86th percentile
  • Coding86th percentile
  • Reasoning88th percentile
  • Multimodal73rd percentile
  • Knowledge89th percentile
  • MultilingualNot eligible
  • Instruction followingNot eligible
  • MathNot eligible

The dashed outline is median of 6 nearest peers.

Eligible category ranks

  1. Agentic#21/140
  2. Coding#23/163
  3. Reasoning#21/161
  4. Multimodal#19/67
  5. Knowledge#20/181
  6. MultilingualNot ranked
  7. Inst. FollowingNot ranked
  8. MathNot ranked
Top decileTop quartileMid-fieldNot eligible

Benchmark ledger

Every result says who ran the test. For this model, Moonshot AI ran 34 of 77. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.

Overall rank
#20
of 218 on the leaderboard
Reported by Moonshot AI
34
Shown with that label
Other publishers
4
Another publisher’s document or table
Run by a benchmark
34
The benchmark’s own run
Independent runs
5
An outside evaluator’s run
Coding18 rows
Coding benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
FrontierSWE v225.9%

Best verified: Claude Fable 5.2 · 69.9%

Gap44 behindRun by FrontierSWE v2Counts8% ref. weight
LiveCodeBench (Vals)LiveCodeBench, Vals AI run87.2%

Best verified: Claude Fable 5.1 · 90.5%

Gap3.3 behindIndependent run: Vals AICounts8% ref. weight
FrontierCode 2.018.9%

Best verified: Claude Fable 5.2 · 50.2%

Gap31.3 behindRun by FrontierCode 2.0Counts4% ref. weight
DeepSWE67.5%

Best verified: Muse Wonder Full · 79.0%

Gap11.5 behindReported by Moonshot AIShown only
ScarfBenchScarfBench: Cross-Framework Application Migration in Enterprise Java61.3%

Best verified: Gemini 4 Argon · 75.2%

Gap13.9 behindRun by ScarfBenchShown only
CursorBench 3.260.8%

Best verified: Claude Fable 5.1 · 73.4%

Gap12.6 behindRun by CursorBench 3.2Shown only
FrontierSWE81.2%

Best verified: Kimi K3 · 81.2%

GapHighest listedReported by Moonshot AIShown only
ProgramBenchProgramBench: Can Language Models Rebuild Programs From Scratch?77.8%

Best verified: Claude Opus 5 · 93.0%

Gap15.2 behindReported by Moonshot AIShown only
Kimi Code Bench v272.9%

Best verified: Kimi K3.1 Pro · 77.1%

Gap4.2 behindReported by Moonshot AIShown only
sweMarathon42%

Best verified: Step 5 Preview · 72.7%

Gap30.7 behindReported by Moonshot AIShown only
PostTrainBench (lab runs)PostTrainBench lab-run results with no stated version36.6%

Best verified: GLM-5.3 · 39.8%

Gap3.2 behindReported by Moonshot AIShown only
MLS-Bench Lite48.3%

Best verified: Kimi K3 · 48.3%

GapHighest listedReported by Moonshot AIShown only
VulcanBench v373.7%

Best verified: Grok 4.5 · 89.9%

Gap16.2 behindRun by VulcanBench v3Shown only
OpenHarmony Bench v1.057.3%

Best verified: Qwen3.8 Max · 60.8%

Gap3.5 behindRun by OpenHarmony Bench v1.0Shown only
SWE-bench (Vals)SWE-bench, Vals AI run93.4%

Best verified: Claude Opus 5 · 97.0%

Gap3.6 behindIndependent run: Vals AIShown only
PostTrainBench v1.132.0%

Best verified: Claude Opus 5.5 · 49.3%

Gap17.3 behindRun by PostTrainBench v1.1Shown only
Bug Hunt Bench21.0 fixes

Best verified: Claude Fable 5.2 · 58.0 fixes

Gap37 behindRun by Bug Hunt BenchShown only
PostTrainBenchPostTrainBench v1.231.5%

Best verified: Claude Opus 5.5 · 43.8%

Gap12.3 behindRun by PostTrainBenchShown only
Agentic19 rows
Agentic benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
Terminal-Bench 4.123.00%

Best verified: Claude Fable 5.2 · 65.80%

Gap42.8 behindRun by Terminal-Bench 4.1Counts9% ref. weight
Terminal-Bench 2.188.3%

Best verified: SWE-2 · 92.8%

Gap4.5 behindReported by Moonshot AICounts9% ref. weight
BrowseComp91.2%

Best verified: Atria Dawn Preview · 92.5%

Gap1.3 behindReported by Moonshot AICounts9% ref. weight
Terminal-Bench 2.1 (Vals)Terminal-Bench 2.1, Vals AI run80.9%

Best verified: Claude Opus 5.5 · 87.6%

Gap6.7 behindIndependent run: Vals AICounts4% ref. weight
AutomationBench30.8%

Best verified: MiniMax M3.1 Flash · 62.2%

Gap31.4 behindReported by Moonshot AIShown only
JobBench52.9%

Best verified: Muse Wonder Full · 75.3%

Gap22.4 behindReported by Moonshot AIShown only
MCP Atlas84.2%

Best verified: Muse Spark Open · 89.2%

Gap5 behindReported by Moonshot AIShown only
Toolathlon-Verified73.2%

Best verified: Claude Opus 5 · 80.6%

Gap7.4 behindReported by Moonshot AIShown only
DeepSearchQA95.0%

Best verified: Atria Dawn Preview · 96.0%

Gap1 behindReported by Moonshot AIShown only
FinchFinch (FinWorkBench): Finance & Accounting across Spreadsheet-Centric Enterprise Workflows77.3%

Best verified: Mistral Large 4 · 73.3%

Gap4 behindReported by a third partyShown only
b3 Agent Securityb3: Backbone Breaker Benchmark88.0%

Best verified: Mistral Large 4 · 97.1%

Gap9.1 behindReported by a third partyShown only
Cyber Misuse RefusalRefusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm)94.0%

Best verified: Mistral Large 4 · 96.6%

Gap2.6 behindReported by a third partyShown only
APEX-Agents37.6%

Best verified: Hy4 · 75.2%

Gap37.6 behindReported by Moonshot AIShown only
SpreadsheetBench 234.8%

Best verified: ERNIE 5.5 Flash Preview · 36.4%

Gap1.6 behindReported by Moonshot AIShown only
DECK-BenchDECK-Bench (Internal)73.5%

Best verified: Kimi K3 · 73.5%

GapHighest listedReported by Moonshot AIShown only
ApprenticeBenchApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job18%

Best verified: Claude Fable 5.1 · 72%

Gap54 behindRun by ApprenticeBenchShown only
Hermes IndexHermes Index: agentic capability and cost per task inside Hermes Agent39.2%

Best verified: Claude Opus 5.5 · 63.3%

Gap24.2 behindRun by Hermes IndexShown only
Hermes BenchHermes Bench: 150 agent tasks graded on the files and state the agent leaves behind71.5%

Best verified: Claude Opus 5.5 · 76.7%

Gap5.2 behindRun by Hermes BenchShown only
MazeBenchMazeBench: visual spatial reasoning in a 3D open world1%

Best verified: GPT-6 Astra · 24%

Gap23 behindRun by MazeBenchShown only
Reasoning2 rows
Reasoning benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
ARC-AGI-2Abstraction and Reasoning Corpus for AGI v260.4%

Best verified: GPT-6.2 Astra · 95.8%

Gap35.4 behindRun by ARC-AGI-2Counts15% ref. weight
ARC-AGI-1ARC-AGI-1 Semi-Private Evaluation94.50%

Best verified: Claude Fable 5 · 98.50%

Gap4 behindRun by ARC-AGI-1Shown only
Multimodal14 rows
Multimodal benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
MMMU-ProMassive Multi-discipline Multimodal Understanding Pro81.6%

Best verified: GPT-6.1 Terra · 85.5%

Gap3.9 behindReported by Moonshot AICounts28% ref. weight
OfficeQA Pro63.3%

Best verified: Claude Fable 5.2 · 72.9%

Gap9.6 behindReported by Moonshot AICounts19% ref. weight
CharXivCharXiv Reasoning91.3%

Best verified: Qwen3.8 Max · 93.5%

Gap2.2 behindReported by Moonshot AICounts16% ref. weight
CharXiv w/o toolsCharXiv Reasoning without tools84.8%

Best verified: Kimi K3.1 Pro · 89%

Gap4.2 behindReported by Moonshot AIShown only
ExtractBench (LlamaIndex)ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)83.17%

Best verified: GPT-6.1 Sol · 95.31%

Gap12.1 behindRun by ExtractBench (LlamaIndex)Shown only
MMMU-Pro w/ PythonMMMU-Pro with Python83.4%

Best verified: Kimi K3.1 Pro · 90%

Gap6.6 behindReported by Moonshot AIShown only
MathVision94.3%

Best verified: Qwen3.8 Max · 95.2%

Gap0.9 behindReported by Moonshot AIShown only
MathVision w/ PythonMathVision with Python97.8%

Best verified: Kimi K3 · 97.8%

GapHighest listedReported by Moonshot AIShown only
BabyVision w/ PythonBabyVision with Python85.7%

Best verified: Kimi K3.1 Pro · 95.4%

Gap9.7 behindReported by Moonshot AIShown only
ZeroBench23.0%

Best verified: Muse Spark Open · 50.0%

Gap27 behindReported by Moonshot AIShown only
ZeroBench w/ PythonZeroBench_main with Python41.0%

Best verified: Kimi K3.1 Pro · 63.0%

Gap22 behindReported by Moonshot AIShown only
WorldVQA ForceAnswer51.0%

Best verified: Kimi K3 · 51.0%

GapHighest listedReported by Moonshot AIShown only
OmniDocBench91.1%

Best verified: Kimi K3 · 91.1%

GapHighest listedReported by Moonshot AIShown only
PerceptionBenchPerceptionBench (Internal)58.5%

Best verified: Kimi K3.1 Pro · 68.3%

Gap9.8 behindReported by Moonshot AIShown only
Knowledge6 rows
Knowledge benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
HLEHumanity's Last Exam56%

Best verified: Claude Fable 5.2 · 74.2%

Gap18.2 behindReported by Moonshot AICounts44% ref. weight
MMLU-Pro (Vals)MMLU-Pro, Vals AI run88.0%

Best verified: Claude Fable 5.1 · 92.4%

Gap4.4 behindIndependent run: Vals AICounts6% ref. weight
GPQAGraduate-Level Google-Proof Q&A93.5%

Best verified: GPT-6 Astra · 96%

Gap2.5 behindReported by Moonshot AICounts3% ref. weight
GPQA Diamond (Vals)GPQA Diamond, Vals AI run92.9%

Best verified: Gemini 3.1 Pro · 95.5%

Gap2.6 behindIndependent run: Vals AICounts2% ref. weight
HLE w/o toolsHumanity's Last Exam without tools43.5%

Best verified: Claude Fable 5.2 · 73.2%

Gap29.7 behindReported by Moonshot AIShown only
GPQA-DGPQA Diamond93.5%

Best verified: GPT-6 Astra · 96.0%

Gap2.5 behindReported by Moonshot AIShown only
Inst. Following1 row
Inst. Following benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
Gray Swan IPI (15 attempts)Gray Swan Indirect Prompt Injection, attack success at 15 attempts52.7%

Best verified: Gemini 4 Carbon · 0.0%

Gap52.7 behindLab-reported, from another publisher’s documentShown only
External signals17 rows
External signals benchmark results, best verified comparison, who ran each test, and whether it counts toward a rank
ASCII Arena1301

Best verified: Gemini 4 Carbon · 2020

Gap719 behindRun by ASCII ArenaShown only
Pelican Arena1496

Best verified: GPT-6 Astra · 1817

Gap321 behindRun by Pelican ArenaShown only
FrontierPhysicsEnd-to-End Physics Research Benchmark for AI Agents2.4%

Best verified: Claude Opus 5.5 · 27.4%

Gap25 behindRun by FrontierPhysicsShown only
NYT ConnectionsNYT Connections LLM Benchmark (extended version)93.6%

Best verified: Gemini 4 Carbon · 99.3%

Gap5.7 behindRun by NYT ConnectionsShown only
Landing Page Arena1564

Best verified: GPT-6.1 Sol · 1695

Gap131 behindRun by Landing Page ArenaShown only
Code2Video1481

Best verified: Claude Opus 5.5 · 1585

Gap104 behindRun by Code2VideoShown only
StealthMarkStealthMark Intelligence Benchmarks (overall score)50.6

Best verified: GPT-6 Astra · 74.6

Gap24 behindRun by StealthMarkShown only
BabaIsBenchBabaIsBench (Baba Is You intro levels, pass@1)96%

Best verified: Claude Opus 5 · 100%

Gap4 behindRun by BabaIsBenchShown only
ExploitBenchExploitBench v8-bench32%

Best verified: GPT-6 Astra · 100%

Gap68 behindRun by ExploitBenchShown only
ACE solvedACE Cyber Range Challenges Solved0%

Best verified: Kimi K3 · 0%

GapBest verifiedRun by ACE solvedShown only
The Last Ones stepsThe Last Ones Average Progress17%

Best verified: Kimi K3 · 17%

GapBest verifiedRun by The Last Ones stepsShown only
The Last Ones completionThe Last Ones Cyber Range Completion Rate10%

Best verified: GPT-5.6 Sol · 70%

Gap60 behindRun by The Last Ones completionShown only
Chess PuzzlesChess Puzzles (Epoch AI)36.0%

Best verified: Claude Fable 5.2 · 79.0%

Gap43 behindRun by Chess PuzzlesShown only
Mystery Game PuzzlesMystery Game Puzzles (Epoch AI)26.0%

Best verified: GPT-6.2 Astra · 95.0%

Gap69 behindRun by Mystery Game PuzzlesShown only
Furniture AssemblyFurniture Assembly Benchmark (Epoch AI)34.2%

Best verified: Claude Fable 5.2 · 90.0%

Gap55.8 behindRun by Furniture AssemblyShown only
Epoch Automation Reports52.4%

Best verified: Claude Fable 5.1 · 65.3%

Gap12.9 behindRun by Epoch Automation ReportsShown only
MirrorCodeMirrorCode (Epoch AI and METR)54.4%

Best verified: Claude Sonnet 5.5 · 97.8%

Gap43.4 behindRun by MirrorCodeShown only

Outlined bars are lab-reported. Filled bars are benchmark-run or independent.

Scores by reasoning effort

Kimi K3 has published results at 3 reasoning effort settings on 3 benchmarks, from ARC Prize and Cognition. Each table is one benchmark run at every effort the publisher tested, with the cost it reported. BenchLM's score uses at most one setting per benchmark; these tables show the rest.

ARC-AGI-1 · Semi-Private set

From Low to Max effort, the score goes from 65.7% to 94.5% and the cost per task from $0.18 to $0.77.

ARC-AGI-1, Semi-Private set: score and cost per task at each reasoning effort
EffortScoreCost per task
Low65.7%$0.18
High86.7%$0.48
Maxtop score94.5%$0.77

Source: ARC Prize.

ARC-AGI-2 · Semi-Private set

From Low to Max effort, the score goes from 12.4% to 60.4% and the cost per task from $0.25 to $1.59.

ARC-AGI-2, Semi-Private set: score and cost per task at each reasoning effort
EffortScoreCost per task
Low12.4%$0.25
High55.0%$0.95
Maxtop score60.4%$1.59

Source: ARC Prize.

FrontierCode 2.0 · Devin CLI

From Low to Max effort, the score goes from 11.3% to 18.4% and the cost per task from $2.98 to $5.46. High scores highest, at 18.9% for $4.47 instead of $5.46.

FrontierCode 2.0, Devin CLI: score and cost per task at each reasoning effort
EffortScoreCost per task
Low11.3%$2.98
Hightop score18.9%$4.47
Max18.4%$5.46

Source: Cognition. Cognition’s mean spend per rollout at list API prices.

Publisher scores and costs · display only, not part of the BenchLM score

How the score is built

What it costs to get this score

Published API price against the public score. The x-axis uses a log scale; the dashed path marks models that are not beaten by a cheaper, higher-scoring option. Price uses average of published input and output rates.

Current modelExplore all models

Kimi K3 · 67.0 score · $9 blended per million tokens

The chart opens on the current model. Scroll horizontally to inspect the full price axis.

405060708090100$0.10$0.50$1$5$10$25$50$100↘ frontierKimi K3

Horizontal: blended price per million tokens, log scale · Vertical: public score

Lineage

The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.

  1. Jan 27, 2026

    Kimi K2.5

    Score 52.1 · $0.6 / $3

  2. Apr 20, 2026

    Kimi K2.6

    Score 56.1 · $0.95 / $4

  3. Jul 16, 2026 · you are here

    Kimi K3

    Score 67.0 · $3 / $15

  4. Oct 23, 2026

    Kimi K3.1 Pro

    Score 71.2 · $2 / $10

Base entry

Spec sheet

Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.

API model ID
Not published
Context window
1.05M
Maximum output
Not sourced yet
Knowledge cutoff
Not sourced yet
Input modalities
text, image, videoKimi-K3 model card
Output modalities
textKimi-K3 model card
Parameters
Not sourced yet
Availability
Not sourced yet
Cloud regions
Not tracked yet
Lifecycle
Superseded
API capabilities
Tool calling, structured outputs, and batch support are not tracked yet
Prompt caching
Published at $0.30 per million cached input tokens
Self-host
Kimi K3 needs ~2304GB VRAM (8× NVIDIA B300 (288GB)).
Rate limits
Not tracked yet

How to read this profile

The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.

Kimi K3 ranks #20 of 218 on the public leaderboard with a score of 67/100. Its source-verified position is #11 of 79.

Kimi K3 is a open weight model with a 1.05M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.

Kimi K3 has cybersecurity benchmark results. The cybersecurity benchmark comparison shows them beside other models, one security job at a time, with the access route for restricted models.

Its explicit predecessor is Kimi K2.6. 77 of 922 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.

Its strongest eligible category is Multimodal & Grounded at #19, while its lowest eligible position is Coding at #23. particularly strong for screenshots, documents, charts, and grounded multimodal workflows.

Last updated November 17, 2026. Runtime fields remain blank until a sourced snapshot exists.

Deployment options

Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.

Where to run Kimi K3 by API

7 hosts publish a per-token price for it. DeepInfra has the lowest output rate, $14.25 per million tokens. Prices read from each host’s own pages on November 16, 2026.

Inference hosts for Kimi K3 with input, output and cached-input prices per million tokens.
HostInput /1MOutput /1MCached /1MContextListed speed
DeepInfra$2.85$14.25$0.281.05M—
Baseten$3.00$15.00$0.30——
Fireworks AI$3.00$15.00$0.30——
Nebius Token Factory$3.00$15.00$0.301.05M264.55 tok/s
Novita AI$3.00$15.00$0.301.05M—
Together AI$3.00$15.00$0.301.05M—
OpenRouter$0.19$16.00$0.181.05M—
Compare inference hosts

Self-host vs API cost

Estimates at 50,000 req/day · 1000 tokens/req average.

Kimi K3
API / mo$13,500
Self-host / mo$57,606
Break-evenNot reached at 70% load
Model the full break-even

Questions

How does Kimi K3 perform overall in AI benchmarks?

Kimi K3 ranks #20 out of 218 models on the public BenchAlign leaderboard, with a score of 67/100. Its evidence status is Supported, and this profile shows 77 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.

Is Kimi K3 good for knowledge and understanding?

Kimi K3 ranks #20 out of 181 eligible models for knowledge and understanding, with a public category score of 67.5/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Kimi K3 good for coding and programming?

Kimi K3 ranks #23 out of 163 eligible models for coding and programming, with a public category score of 56.3/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Kimi K3 good for reasoning and logic?

Kimi K3 ranks #21 out of 161 eligible models for reasoning and logic, with a public category score of 65.4/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

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

Kimi K3 ranks #21 out of 140 eligible models for agentic tool use and computer tasks, with a public category score of 65.1/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Kimi K3 good for multimodal and grounded tasks?

Kimi K3 ranks #19 out of 67 eligible models for multimodal and grounded tasks, with a public category score of 66.7/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.

Is Kimi K3 good for instruction following?

Kimi K3 has source-displayable benchmark coverage for instruction following, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.

Is Kimi K3 open source?

Kimi K3 is an open-weight model from Moonshot AI. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.

Does Kimi K3 have full benchmark coverage on BenchLM?

No. Kimi K3 currently has 102 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.

What is the context window size of Kimi K3?

Kimi K3 has a reported context window of 1.05M in the exact-model catalog record. The value stays visible, but the profile marks its source link as unavailable instead of presenting it as directly documented. Maximum output length remains separate because providers often publish a different limit.

Compare Kimi K3 with every tracked model1041 comparisons