Claude Opus 4.6

Leaderboards

Where Claude Opus 4.6 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 Claude Opus 4.6 has a published result on, out of 100%, with its rank among the models that took it.

• GPQA

  #21 / 256

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

  91.3%

  Verified

• AIME 2025

  #6 / 123

  Competition maths problems from the 2025 American Invitational Mathematics Exam.

  99.8%

  Self-reported

• SWE-Bench Verified

  #8 / 120

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

  80.8%

  Self-reported

• Humanity's Last Exam

  #18 / 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

  53.1%

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

• MMMU-Pro

  #24 / 73

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

  77.3%

  Self-reported · In the vision score

• BrowseComp

  #16 / 68

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

  84.0%

  Self-reported · In the agents score

• CharXiv-R

  #37 / 60

  Reading and reasoning about charts from scientific papers.

  77.4%

  Self-reported

• Terminal-Bench 2.0

  #17 / 54

  Getting real work done in a command-line terminal.

  65.4%

  Self-reported

• MMMLU

  #6 / 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.

  91.1%

  Self-reported

• SWE-bench Multilingual

  #12 / 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.

  77.8%

  Self-reported

• LiveBench

  #11 / 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.

  76.3%

  Self-reported · In the text score

• MCP Atlas

  #30 / 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.

  62.7%

  Self-reported

• Tau2 Telecom

  #1 / 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.

  99.3%

  Self-reported

• MRCR v2 (8-needle)

  #6 / 32

  MRCR v2 (8-needle) is a variant of the Multi-Round Coreference Resolution benchmark that includes 8 needle items to retrieve from long contexts. This tests models' ability to simultaneously track and reason about multiple pieces of information across extended conversations.

  76.0%

  Self-reported · In the text score

• Tau2 Retail

  #1 / 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.

  91.9%

  Self-reported

• CyberGym

  #16 / 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.

  73.8%

  Self-reported

• OSWorld

  #3 / 21

  OSWorld: The first-of-its-kind scalable, real computer environment for multimodal agents, supporting task setup, execution-based evaluation, and interactive learning across Ubuntu, Windows, and macOS with 369 computer tasks involving real web and desktop applications, OS file I/O, and multi-application workflows

  72.7%

  Self-reported

• ARC-AGI v2

  #6 / 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.

  68.8%

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

• FrontierSWE

  #9 / 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.

  56.0%

  Self-reported

• Legal Agent Benchmark

  #8 / 15

  The Legal Agent Benchmark (LAB) is Harvey's open-source benchmark for evaluating AI agents on complex, long-horizon legal work. Tasks are scored under an all-pass standard against expert-curated rubrics, where a task passes only if every required rubric criterion (facts, conclusions, citations, structure, and analytical moves) passes.

  4.2%

  Self-reported

• Graphwalks BFS >128k

  #7 / 12

  A graph reasoning benchmark that evaluates language models' ability to perform breadth-first search (BFS) operations on graphs with context length over 128k tokens, testing long-context reasoning capabilities.

  61.5%

  Self-reported

• DeepSearchQA

  #3 / 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.

  91.3%

  Self-reported

• GPQA (diamond)

  #5 / 10

  Graduate-level Google-Proof Q&A benchmark with paper-aligned full and diamond subsets.

  84.8%

  Unverified

• Finance Agent

  #3 / 8

  Finance Agent is a benchmark for evaluating AI models on agentic financial analysis tasks, testing their ability to process financial data, perform calculations, and generate accurate analyses across various financial domains.

  60.7%

  Self-reported

• Graphwalks parents >128k

  #1 / 7

  A graph reasoning benchmark that evaluates language models' ability to find parent nodes in graphs with context length over 128k tokens, testing long-context reasoning and graph structure understanding.

  95.4%

  Self-reported

• Vending-Bench 2

  #1 / 4

  Vending-Bench 2 tests longer horizon planning capabilities by evaluating how well AI models can manage a simulated vending machine business over extended periods. The benchmark measures a model's ability to maintain consistent tool usage and decision-making for a full simulated year of operation, driving higher returns without drifting off task.

  801759.0%

  Self-reported

• FigQA

  #2 / 3

  FigQA is a multiple-choice benchmark on interpreting scientific figures from biology papers. It evaluates dual-use biological knowledge and multimodal reasoning relevant to bioweapons development.

  78.3%

  Self-reported

• OpenRCA

  #1 / 1

  OpenRCA is a benchmark for evaluating AI models on root cause analysis tasks. For each failure case, the model receives 1 point if all generated root-cause elements match the ground-truth ones, and 0 points if any mismatch is identified. The overall accuracy is the average score across all failure cases.

  34.9%

  Unverified