GPT-5.2

Leaderboards

Where GPT-5.2 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 GPT-5.2 has a published result on, out of 100%, with its rank among the models that took it.

• GPQA

  #14 / 256

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

  92.4%

  Self-reported

• AIME 2025

  #1 / 123

  Competition maths problems from the 2025 American Invitational Mathematics Exam.

  100.0%

  Self-reported

• SWE-Bench Verified

  #16 / 120

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

  80.0%

  Self-reported

• Humanity's Last Exam

  #50 / 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

  34.5%

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

• MMMU-Pro

  #16 / 73

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

  79.5%

  Self-reported · In the vision score

• BrowseComp

  #37 / 68

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

  65.8%

  Self-reported · In the agents score

• CharXiv-R

  #21 / 60

  Reading and reasoning about charts from scientific papers.

  82.1%

  Self-reported

• MMMLU

  #9 / 51

  Multilingual Massive Multitask Language Understanding dataset released by OpenAI, featuring professionally translated MMLU test questions across 14 languages including Arabic, Bengali, German, Spanish, French, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Swahili, Yoruba, and Chinese. Contains approximately 15,908 multiple-choice questions per language covering 57 subjects.

  89.6%

  Self-reported

• Toolathlon

  #29 / 42

  Tool Decathlon is a comprehensive benchmark for evaluating AI agents' ability to use multiple tools across diverse task categories. It measures proficiency in tool selection, sequencing, and execution across ten different tool-use scenarios.

  46.3%

  Self-reported · In the agents score

• LiveBench

  #16 / 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.

  74.8%

  Self-reported · In the text score

• MCP Atlas

  #33 / 38

  MCP Atlas is a benchmark for evaluating AI models on scaled tool use capabilities, measuring how well models can coordinate and utilize multiple tools across complex multi-step tasks.

  60.6%

  Self-reported

• Tau2 Telecom

  #4 / 36

  τ²-Bench telecom domain evaluates conversational agents in a dual-control environment modeled as a Dec-POMDP, where both agent and user use tools in shared telecommunications troubleshooting scenarios that test coordination and communication capabilities.

  98.7%

  Self-reported

• HMMT 2025

  #2 / 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

  99.4%

  Self-reported

• VideoMMMU

  #4 / 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.

  85.9%

  Self-reported

• ScreenSpot Pro

  #4 / 27

  ScreenSpot-Pro is a novel GUI grounding benchmark designed to rigorously evaluate the grounding capabilities of multimodal large language models (MLLMs) in professional high-resolution computing environments. The benchmark comprises 1,581 instructions across 23 applications spanning 5 industries and 3 operating systems, featuring authentic high-resolution images from professional domains with expert annotations. Unlike previous benchmarks that focus on cropped screenshots in consumer applications, ScreenSpot-Pro addresses the complexity and diversity of real-world professional software scenarios, revealing significant performance gaps in current MLLM GUI perception capabilities.

  86.3%

  Self-reported

• Tau2 Retail

  #6 / 27

  τ²-bench retail domain evaluates conversational AI agents in customer service scenarios within a dual-control environment where both agent and user can interact with tools. Tests tool-agent-user interaction, rule adherence, and task consistency in retail customer support contexts.

  82.0%

  Self-reported

• ARC-AGI v2

  #9 / 19

  ARC-AGI-2 is an upgraded benchmark for measuring abstract reasoning and problem-solving abilities in AI systems through visual grid transformation tasks. It evaluates fluid intelligence via input-output grid pairs (1x1 to 30x30) using colored cells (0-9), requiring models to identify underlying transformation rules from demonstration examples and apply them to test cases. Designed to be easy for humans but challenging for AI, focusing on core cognitive abilities like spatial reasoning, pattern recognition, and compositional generalization.

  52.9%

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

• FrontierMath

  #5 / 17

  A benchmark of hundreds of original, exceptionally challenging mathematics problems crafted and vetted by expert mathematicians, covering most major branches of modern mathematics from number theory and real analysis to algebraic geometry and category theory.

  40.3%

  Self-reported

• Graphwalks BFS <128k

  #2 / 13

  A graph reasoning benchmark that evaluates language models' ability to perform breadth-first search (BFS) operations on graphs with context length under 128k tokens, returning nodes reachable at specified depths.

  94.0%

  Self-reported

• Graphwalks parents <128k

  #3 / 12

  A graph reasoning benchmark that evaluates language models' ability to find parent nodes in graphs with context length under 128k tokens, requiring understanding of graph structure and edge relationships.

  89.0%

  Self-reported

• ARC-AGI

  #6 / 11

  The Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI) is a benchmark designed to test general intelligence and abstract reasoning capabilities through visual grid-based transformation tasks. Each task consists of 2-5 demonstration pairs showing input grids transformed into output grids according to underlying rules, with test-takers required to infer these rules and apply them to novel test inputs. The benchmark uses colored grids (up to 30x30) with 10 discrete colors/symbols, designed to measure human-like general fluid intelligence and skill-acquisition efficiency with minimal prior knowledge.

  86.2%

  Self-reported

• SWE-Lancer (IC-Diamond subset)

  #3 / 6

  SWE-Lancer (IC-Diamond subset) is a benchmark of real-world freelance software engineering tasks from Upwork, ranging from $50 bug fixes to $32,000 feature implementations. It evaluates AI models on independent engineering tasks using end-to-end tests triple-verified by experienced software engineers, and includes managerial tasks where models choose between technical implementation proposals.

  74.6%

  Self-reported

• BrowseComp Long Context 128k

  #1 / 5

  A challenging benchmark for evaluating web browsing agents' ability to persistently navigate the internet and find hard-to-locate, entangled information. Comprises 1,266 questions requiring strategic reasoning, creative search, and interpretation of retrieved content, with short and easily verifiable answers.

  92.0%

  Self-reported

• BrowseComp Long Context 256k

  #1 / 2

  BrowseComp is a benchmark for measuring the ability of agents to browse the web, comprising 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers. The benchmark focuses on questions where answers are obscure, time-invariant, and well-supported by evidence scattered across the open web.

  89.8%

  Unverified