The LLM benchmark aggregator.

Kimi K2.5

Moonshot · 1T · open weights

Strongest in Core abilities (#16 of 199, on 1 of its 4 boards), weakest in Safety (#216 of 329). Above par in 17 of 26 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5.5 and Gemini 3 Pro and ahead of GPT-6 Luna and Claude 3.5 Sonnet.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Core abilities58.8−6.5#16/1991/4

General intelligence62.7−6.7#17/1781/3

Human preference61.5−7.3#65/3541/1

Knowledge52.4−10.1#79/1971/3

Academic knowledge55.9−8#51/1231/1

Agents53−13.4#80/2982/6

Computer use57.8−10.3#18/681/2

Knowledge work50.2−20.5#94/2081/2

Reasoning51.8−13.6#84/1953/5

Mathematics57.8−6#24/1521/2

Science55.6−7.9#55/1221/1

Reasoning45−24#89/1472/3

Professional50.2−12.2#94/1941/2

Finance53.3−9.1#71/1621/1

Medical50.6−15.9#81/1491/1

Legal44.5−21.4#124/1731/1

Coding42−24.2#156/1791/6

Agentic coding41.6−25.3#148/1731/5

Safety48.7−12.7#216/3292/3

Secure code63.4−2.8#24/2701/1

Factual grounding39.2−31.5#90/1041/1

Toxicity avoidance53.7−5.3#132/2681/1

Fairness46.2−23.8#160/2921/2

Harm refusal48.9−12.6#168/2921/2

Jailbreak resistance40.6−24.4#211/2681/1

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Kimi K2.5, left for the other.

Mistral Medium 3.5won 17 · lost 0

Grok 4.3won 20 · lost 2

Mistral Large 3won 18 · lost 2

Magistral Small 1.2won 15 · lost 2

Command R+won 15 · lost 2

Command Awon 20 · lost 3

GPT OSS 20Bwon 18 · lost 3

NVIDIA Nemotron 3 Ultra 550B A55Bwon 20 · lost 4

GPT-5.4 nanowon 16 · lost 4

DeepSeek V3won 20 · lost 5

GPT-4.1 nanowon 19 · lost 5

Gemini 3.1 Flash-Litewon 15 · lost 4

GPT-4o miniwon 18 · lost 5

Llama 4 Scout Instructwon 18 · lost 5

Ling 3.0 Flashwon 14 · lost 4

Grok 4.1 Fastwon 17 · lost 5

Gemini 1.5 Flashwon 12 · lost 4

DeepSeek V3.1won 12 · lost 4

DeepSeek R1won 18 · lost 6

GLM-4.5won 14 · lost 5

GPT-5.4 miniwon 14 · lost 5

Gemma 4 31Bwon 16 · lost 6

Gemini 1.5 Prowon 13 · lost 5

Gemini 2.5 Flash-Litewon 13 · lost 5

Grok 4 Fastwon 13 · lost 5

GPT OSS 120Bwon 17 · lost 7

Kimi K2won 17 · lost 7

Llama 3.3 70B Instructwon 14 · lost 6

Grok 3won 16 · lost 7

GLM-4.6won 16 · lost 7

Gemini 2.0 Flashwon 15 · lost 7

DeepSeek V3.2won 17 · lost 8

MiniMax M2.5won 12 · lost 6

Inkling Smallwon 12 · lost 6

Gemini 3.5 Flash-Litewon 12 · lost 6

GPT-4.1 miniwon 16 · lost 8

Claude Sonnet 4won 17 · lost 9

Claude Haiku 4.5won 15 · lost 8

Claude 3.5 Haikuwon 13 · lost 7

Llama 4 Maverick Instructwon 16 · lost 9

GPT-5 nanowon 14 · lost 8

O3 Miniwon 14 · lost 8

MiniMax M2.1won 10 · lost 6

O4 Miniwon 15 · lost 9

Gemini 2.5 Flashwon 15 · lost 9

MiMo V2.5 Prowon 13 · lost 8

GLM-4.7won 11 · lost 7

Grok 4won 14 · lost 9

Qwen3 235B A22B Instructwon 14 · lost 10

Grok 3 Miniwon 11 · lost 8

Qwen3.8 27Bwon 11 · lost 8

GPT-4owon 13 · lost 10

Qwen3.6 Pluswon 9 · lost 7

GPT-5 miniwon 14 · lost 11

Gemini 2.5 Prowon 14 · lost 11

Kimi K2 Thinkingwon 11 · lost 9

MiMo V2.6 Flashwon 11 · lost 9

Mistral Large 4won 12 · lost 10

DeepSeek V4 Flashwon 13 · lost 11

Claude 3.5 Sonnetwon 10 · lost 9

GPT-6 Lunawon 12 · lost 11

Grok 4.20won 8 · lost 8

Claude 3.7 Sonnetwon 11 · lost 11

GPT-4.1won 12 · lost 12

O3won 12 · lost 12

MiniMax M3won 12 · lost 13

MiniMax M2.7won 10 · lost 12

O1won 9 · lost 11

Inklingwon 10 · lost 15

Claude Opus 4won 9 · lost 14

Qwen3.7 Maxwon 9 · lost 14

Kimi K2.6won 9 · lost 15

Grok 4.7won 8 · lost 14

MiMo V2.6 Prowon 8 · lost 14

GLM-5.3 Flashwon 9 · lost 16

Claude Haiku 5.5won 6 · lost 11

GLM-5.2won 8 · lost 16

GLM-5.1won 7 · lost 15

Gemini 3.6 Flashwon 7 · lost 15

Gemini 3.5 Flashwon 7 · lost 16

GPT-5won 7 · lost 18

Gemini 3.8 Flashwon 7 · lost 18

GLM-5won 5 · lost 13

Claude Sonnet 4.5won 7 · lost 19

GPT-5.6 Lunawon 6 · lost 18

DeepSeek V4 Prowon 6 · lost 19

DeepSeek V4.1 Flashwon 4 · lost 15

Kimi K3won 4 · lost 15

GPT-5.6 Terrawon 5 · lost 20

Qwen3.8 Maxwon 5 · lost 20

Claude Opus 4.6won 4 · lost 17

GPT-6 Solwon 4 · lost 18

GPT-5.1won 4 · lost 19

Muse Spark 1.1won 4 · lost 19

GLM-5.3won 3 · lost 15

GPT-5.2won 4 · lost 20

Grok 4.6won 4 · lost 21

GPT-5.6 Solwon 4 · lost 21

Gemini 3.1 Prowon 3 · lost 16

Claude Sonnet 5won 3 · lost 20

Claude Sonnet 4.6won 3 · lost 20

Gemini 3 Flashwon 2 · lost 14

Muse Spark 1.2won 2 · lost 14

Grok 4.5won 3 · lost 21

GPT-5.4won 3 · lost 21

Claude Opus 4.5won 2 · lost 17

Claude Opus 4.8won 2 · lost 22

Claude Opus 4.7won 2 · lost 23

Claude Sonnet 5.5won 1 · lost 16

GPT-6.1 Solwon 1 · lost 16

Gemini 3.7 Flashwon 1 · lost 16

Gemini 4 Argonwon 1 · lost 17

Claude Fable 5.1won 1 · lost 18

Claude Opus 5won 1 · lost 23

GPT-5.5won 1 · lost 24

GPT-6 Astrawon 0 · lost 19

Claude Fable 5won 0 · lost 19

Gemini 3 Prowon 0 · lost 21

Claude Opus 5.5won 0 · lost 24

§ 3 · Sources

Where the numbers come from

9 publications, 28 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-10

LMArena Text 1450

GDPval-AA ↗1 measureread 2026-11-10

GDPval-AA 847

ARC-AGI-2 ↗1 measureread 2026-11-10

ARC-AGI-2 11.8%

OSWorld ↗1 measureread 2026-11-10

OSWorld 63.3%

Kagi LLM Benchmark ↗1 measureread 2026-11-10

Kagi LLM Benchmark 78.5%

SimpleBench ↗1 measureread 2026-11-10

SimpleBench 46.8%

Vals.ai ↗15 measuresread 2026-11-10

Vals · Legal Research Bench 15.87%Vals · CaseLaw 58.73%Vals · Finance Agent 35.79%Vals · CorpFin 68.26%Vals · TaxEval 74.2%Vals · MortgageTax 66.53%Vals · MedQA 94.37%Vals · MedCode 39.32%Vals · MedScribe 76.44%Vals · SWE-bench Verified 70% (Mini-SWE-agent)Vals · Vibe Code Bench 17.54% (OpenHands)Vals · Code Migration 6.96%Vals · GPQA Diamond 84.09%Vals · MMLU Pro 85.91%Vals · AIME 95.63%

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-11-10

Enkrypt · Jailbreak risk 20.7%Enkrypt · Harmful content risk 5.6%Enkrypt · CBRN risk 33.2%Enkrypt · Toxicity risk 3.6%Enkrypt · Bias risk 83.7%Enkrypt · Insecure code risk 5.8%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10

Vectara · Factual consistency 85.8%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=kimi-k2-5)](https://publicai.io/model-index/m/kimi-k2-5)