Kimi K2.5

Leaderboards

Where Kimi K2.5 sits among the top 20 in each category.

By task

Scored from votes on each kind of prompt alone, on the same 1,000-point scale.

Test results

Every test Kimi K2.5 has a published result on, out of 100%, with its rank among the models that took it.

• GPQA

  #49 / 256

  Graduate-level science questions written to be hard to look up.

  87.6%

  Self-reported

• MMLU-Pro

  #8 / 142

  Broad knowledge and reasoning across 14 subjects, ten answer choices each.

  87.1%

  Self-reported

• AIME 2025

  #19 / 123

  Competition maths problems from the 2025 American Invitational Mathematics Exam.

  96.1%

  Self-reported

• SWE-Bench Verified

  #31 / 120

  Fixing real bugs in real GitHub projects, checked by the projects' own tests.

  76.8%

  Self-reported

• Humanity's Last Exam

  #24 / 108

  Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions

  50.2%

  Self-reported · In the text score· In the vision score

• MMMU-Pro

  #19 / 73

  College-level questions that need both an image and text to answer.

  78.5%

  Self-reported · In the vision score

• BrowseComp

  #29 / 68

  Finding hard-to-locate facts by browsing the web.

  74.9%

  Self-reported · In the agents score

• LiveCodeBench v6

  #15 / 67

  LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code. It continuously collects new problems from programming contests (LeetCode, AtCoder, CodeForces) and evaluates four different scenarios: code generation, self-repair, code execution, and test output prediction. Problems are annotated with release dates to enable evaluation on unseen problems released after a model's training cutoff.

  85.0%

  Self-reported

• SWE-Bench Pro

  #56 / 61

  Harder, longer software engineering tasks from real codebases.

  50.7%

  Self-reported · In the coding score

• CharXiv-R

  #35 / 60

  Reading and reasoning about charts from scientific papers.

  77.5%

  Self-reported

• Terminal-Bench 2.0

  #36 / 54

  Getting real work done in a command-line terminal.

  50.8%

  Self-reported

• SWE-bench Multilingual

  #20 / 49

  A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++. Contains 1,632 high-quality instances carefully annotated from 2,456 candidates by 68 expert annotators, designed to evaluate Large Language Models across diverse software ecosystems beyond Python.

  73.0%

  Self-reported

• LiveBench

  #32 / 38

  LiveBench is a challenging, contamination-limited LLM benchmark that addresses test set contamination by releasing new questions monthly based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses. It comprises tasks across math, coding, reasoning, language, instruction following, and data analysis with verifiable, objective ground-truth answers.

  69.1%

  Self-reported · In the text score

• MathVision

  #13 / 38

  MATH-Vision is a dataset designed to measure multimodal mathematical reasoning capabilities. It focuses on evaluating how well models can solve mathematical problems that require both visual understanding and mathematical reasoning, bridging the gap between visual and mathematical domains.

  84.2%

  Self-reported

• HMMT 2025

  #6 / 33

  Harvard-MIT Mathematics Tournament 2025 - A prestigious student-organized mathematics competition for high school students featuring two tournaments (November 2025 at MIT and February 2026 at Harvard) with individual tests, team rounds, and guts rounds

  95.4%

  Self-reported

• LVBench

  #10 / 31

  LVBench is an extreme long video understanding benchmark designed to evaluate multimodal models on videos up to two hours in duration. It contains 6 major categories and 21 subcategories, with videos averaging five times longer than existing datasets. The benchmark addresses applications requiring comprehension of extremely long videos.

  75.9%

  Self-reported

• SciCode

  #9 / 30

  SciCode is a research coding benchmark curated by scientists that challenges language models to code solutions for scientific problems. It contains 338 subproblems decomposed from 80 challenging main problems across 16 natural science sub-fields including mathematics, physics, chemistry, biology, and materials science. Problems require knowledge recall, reasoning, and code synthesis skills.

  48.7%

  Self-reported

• VideoMMMU

  #3 / 28

  Video-MMMU evaluates Large Multimodal Models' ability to acquire knowledge from expert-level professional videos across six disciplines through three cognitive stages: perception, comprehension, and adaptation. Contains 300 videos and 900 human-annotated questions spanning Art, Business, Science, Medicine, Humanities, and Engineering.

  86.6%

  Self-reported

• MathVista-Mini

  #1 / 26

  MathVista-Mini is a smaller version of the MathVista benchmark that evaluates mathematical reasoning in visual contexts. It consists of examples derived from multimodal datasets involving mathematics, combining challenges from diverse mathematical and visual tasks to assess foundation models' ability to solve problems requiring both visual understanding and mathematical reasoning.

  90.1%

  Self-reported

• AA-LCR

  #7 / 25

  Agent Arena Long Context Reasoning benchmark

  70.0%

  Self-reported

• HealthBench Hard

  #11 / 24

  A challenging variation of HealthBench that evaluates large language models' performance and safety in healthcare through 5,000 multi-turn conversations with particularly rigorous evaluation criteria validated by 262 physicians from 60 countries

  24.0%

  Verified

• LongBench v2

  #6 / 24

  LongBench v2 is a benchmark designed to assess the ability of LLMs to handle long-context problems requiring deep understanding and reasoning across real-world multitasks. It consists of 503 challenging multiple-choice questions with contexts ranging from 8k to 2M words across six major task categories: single-document QA, multi-document QA, long in-context learning, long-dialogue history understanding, code repository understanding, and long structured data understanding.

  61.0%

  Self-reported

• OCRBench

  #1 / 24

  OCRBench: Comprehensive evaluation benchmark for assessing Optical Character Recognition (OCR) capabilities in Large Multimodal Models across text recognition, scene text VQA, and document understanding tasks

  92.3%

  Self-reported

• IMO-AnswerBench

  #16 / 22

  IMO-AnswerBench is a benchmark for evaluating mathematical reasoning capabilities on International Mathematical Olympiad (IMO) problems, focusing on answer generation and verification.

  81.8%

  Self-reported

• CyberGym

  #21 / 21

  CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.

  41.3%

  Self-reported

• OmniDocBench 1.5

  #10 / 20

  OmniDocBench 1.5 is a comprehensive benchmark for evaluating multimodal large language models on document understanding tasks, including OCR, document parsing, information extraction, and visual question answering across diverse document types. Lower Overall Edit Distance scores are better.

  88.8%

  Self-reported

• SimpleVQA

  #7 / 17

  SimpleVQA is a visual question answering benchmark focused on simple queries.

  0.7%

  Self-reported

• Video-MME

  #5 / 17

  Video-MME is the first-ever comprehensive evaluation benchmark of Multi-modal Large Language Models (MLLMs) in video analysis. It features 900 videos totaling 254 hours with 2,700 human-annotated question-answer pairs across 6 primary visual domains (Knowledge, Film & Television, Sports Competition, Life Record, Multilingual, and others) and 30 subfields. The benchmark evaluates models across diverse temporal dimensions (11 seconds to 1 hour), integrates multi-modal inputs including video frames, subtitles, and audio, and uses rigorous manual labeling by expert annotators for precise assessment.

  87.4%

  Self-reported

• FrontierSWE

  #15 / 16

  FrontierSWE measures whether an agent can complete open-ended technical projects at the scale of hours to tens of hours, spanning systems optimization, large-scale code construction, and applied ML research. Performance is reported as a dominance score, where higher is better.

  26.0%

  Self-reported

• ZEROBench

  #9 / 15

  ZEROBench is a challenging vision benchmark designed to test models on zero-shot visual understanding tasks.

  0.1%

  Self-reported

• WideSearch

  #4 / 14

  WideSearch is an agentic search benchmark that evaluates models' ability to perform broad, parallel search operations across multiple sources. It tests wide-coverage information retrieval and synthesis capabilities.

  79.0%

  Self-reported

• HealthBench (hard)

  #4 / 13

  24.0%

  Unverified

• InfoVQAtest

  #1 / 12

  InfoVQA test set with infographic images requiring joint reasoning over document layout, textual content, graphical elements, and data visualizations with elementary reasoning and arithmetic skills

  92.6%

  Self-reported

• DeepSearchQA

  #8 / 10

  DeepSearchQA is a benchmark for evaluating deep search and question-answering capabilities, testing models' ability to perform multi-hop reasoning and information retrieval across complex knowledge domains.

  77.1%

  Self-reported

• MMVU

  #2 / 7

  MMVU (Multimodal Multi-disciplinary Video Understanding) is a benchmark for evaluating multimodal models on video understanding tasks across multiple disciplines, testing comprehension and reasoning capabilities on video content.

  80.4%

  Self-reported

• Seal-0

  #1 / 7

  Seal-0 is a benchmark for evaluating agentic search capabilities, testing models' ability to navigate and retrieve information using tools.

  57.4%

  Self-reported

• WorldVQA

  #7 / 7

  WorldVQA is a benchmark designed to evaluate atomic vision-centric world knowledge. It assesses models' ability to understand and reason about visual elements representing real-world knowledge.

  46.3%

  Self-reported

• LongVideoBench

  #3 / 6

  LongVideoBench is a question-answering benchmark featuring video-language interleaved inputs up to an hour long. It includes 3,763 varying-length web-collected videos with subtitles across diverse themes and 6,678 human-annotated multiple-choice questions in 17 fine-grained categories for comprehensive evaluation of long-term multimodal understanding.

  79.8%

  Self-reported

• MotionBench

  #4 / 5

  MotionBench is a benchmark for evaluating multimodal models on motion understanding in videos, testing the ability to comprehend temporal dynamics, movement patterns, and action sequences.

  70.4%

  Self-reported

• nolima (hard_4k)

  #1 / 5

  99.5%

  Independent

• PaperBench

  #2 / 3

  PaperBench is a benchmark for evaluating AI agents on their ability to replicate research papers. It tests models on complex, multi-step workflows involving code implementation, experimentation, and reproducing scientific results from academic publications.

  63.5%

  Self-reported

• independence-bench (default)

  #3 / 3

  50.0%

  Unverified

• FinSearchComp T2&T3

  #1 / 1

  FinSearchComp T2&T3 is a combined benchmark for evaluating financial search and reasoning capabilities on Tier 2 and Tier 3 tasks, testing models' ability to retrieve and analyze complex financial information using tools.

  67.8%

  Unverified

• OJBench (C++)

  #1 / 1

  OJBench (C++) is the C++ subset of OJBench, a competition-level code benchmark that evaluates large language models on programming competition problems from NOI and ICPC using C++ as the implementation language.

  57.4%

  Unverified