Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Anthropic · Proprietary · rank 16 · bench-align-v5
Claude Opus 5 is built for coding, agent work and reasoning. It ranks in the top quarter for coding (top 1% of 210 scored models) and agent work (top 2% of 127 scored models), by BenchLM's published scores.
Every published price component, including cache reads and writes.
Lifecycle events the source associates with this model, plus what this profile does not claim.
Claude Opus 5 by Anthropic scores 74.9 overall, costs $5.00 per 1M input tokens and $25.00 per 1M output tokens, and carries a context window of 1M tokens — observed 2026-11-21.
Seven axes from the ranking source. A missing axis is drawn as a gap.
Each axis shows the ranking source’s score and, beside it, BenchLM’s separate verified score for that axis. The two are published side by side and are not interchangeable.
Measured values with the date they were observed. Nothing is inferred from a sibling model or a provider claim.
Time to first token: latency to the first answer chunk (Artificial Analysis, via BenchLM). Reasoning models include thinking time, so values can run to tens or hundreds of seconds.
Evidence key: Observed
| Measurement | Value | Observed | Last good | Evidence |
|---|---|---|---|---|
| Time to first token | 59.73 s | 2026-11-21 | 2026-11-21 | |
| Throughput | 62 tok/s | 2026-11-21 | 2026-11-21 |
Regional or per-endpoint measurements appear only when the API supplies them; none are modelled here.
| $12.50 |
| Batch cache read / 1M tokens (OpenRouter batch) | $0.25 |
| Batch cache write / 1M tokens (OpenRouter batch) | $3.13 |
Batch API rate (asynchronous jobs), from OpenRouter's batch listing.
10 conversations per day × 8 messages × 22 active days, 1200 input and 400 output tokens per message, no cache. This uses the same calculator as the cost simulator, so an unavailable applicable rate makes the example unavailable too.
Plans from the model's maker, and the API providers that list this exact model today. Links leave ModelChoice; each provider sets its own prices and limits.
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| HealthBench (raw) HealthBench raw score | 67.1 | #3 | 5,000 multi-turn patient conversations | Realistic healthcare conversations | HealthBench: Evaluating Large Language Models Towards Improved Human Health |
| HealthBench (length-adjusted) HealthBench length-adjusted score | 57.8 | #5 | 5,000 multi-turn patient conversations | Realistic healthcare conversations | HealthBench: Evaluating Large Language Models Towards Improved Human Health |
| HealthBench Professional (raw) HealthBench Professional raw score | 73.4 | #3 | 525 physician-authored conversations | Professional clinical tasks | HealthBench Professional: Evaluating Large Language Models on Real Clinician Chats |
| BioMysteryBench (human-solvable) BioMysteryBench Human Solvable | 90.1 | #1 | Human-solvable computational biology investigations | Expert computational biology | Claude Opus 5 System Card |
| BioMysteryBench (human-difficult) BioMysteryBench Human Difficult | 49.4 | #3 | Human-difficult computational biology investigations | Frontier computational biology | Claude Opus 5 System Card |
| SpatialBench Verified LatchBio SpatialBench Verified | 72.5 | #1 | 115 externally validated spatial transcriptomics problems | Professional bioinformatics | Claude Opus 5 System Card |
| SingleCellBench LatchBio SingleCellBench | 60.6 | #4 | 195 single-cell RNA sequencing problems | Professional bioinformatics | Claude Opus 5 System Card |
| ProteinGym Hard ProteinGym Hard | 47.7 | #1 | Hard protein mutation-effect ranking tasks | Computational protein science | Claude Opus 5 System Card |
| Protein Design Anthropic Protein Design evaluation | 42.5 | #4 | Constrained protein-sequence design tasks | Computational protein design | Claude Opus 5 System Card |
| Organic chemistry V2 Anthropic Organic Chemistry V2 evaluation | 61.6 | #1 | Organic chemistry reasoning tasks | Expert organic chemistry | Claude Opus 5 System Card |
| Protocols (troubleshooting) Molecular Biology Protocols Troubleshooting | 61.1 | #5 | Molecular-biology protocol troubleshooting | Expert laboratory protocols | Claude Opus 5 System Card |
| Protocols (understanding) Benchling Molecular Biology Protocols Understanding | 78.4 | #1 | Molecular-biology protocol extension | Expert laboratory protocols | Claude Opus 5 System Card |
| HLE Humanity's Last Exam | 64.7 | #5 | Expert-level questions | Frontier expert level | Humanity's Last Exam |
| HLE-Verified HLE-Verified | 54.4 | #3 | 1,811 verified or revised expert questions | Frontier multidisciplinary expert reasoning | HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam |
| LABBench2 LABBench2: An Improved Benchmark for AI Systems Performing Biology Research | 84.2 | #3 | Nearly 1,900 biology-research tasks | Real-world biology research | LABBench2: An Improved Benchmark for AI Systems Performing Biology Research |
| Artificial Analysis Intelligence Index Artificial Analysis Intelligence Index | 50.8 | #17 | Cross-benchmark intelligence index | Display-only external reference | Artificial Analysis |
| AA-GPQA Diamond Artificial Analysis GPQA Diamond | 93.2 | #11 | Graduate-level science questions | Graduate-level science reasoning | Artificial Analysis GPQA Diamond Benchmark Leaderboard |
| AA-HLE Artificial Analysis Humanity's Last Exam | 54.9 | #7 | Expert-level questions | Frontier expert reasoning | Artificial Analysis Humanity's Last Exam Benchmark Leaderboard |
| AA-Omniscience Index Artificial Analysis Omniscience Index | 37.1 | #13 | Knowledge questions | Broad factual knowledge | AA-Omniscience: Knowledge and Hallucination Benchmark |
| AA-Omniscience Accuracy Artificial Analysis Omniscience Accuracy | 60.9 | #9 | Knowledge questions | Broad knowledge | Artificial Analysis model benchmarks |
| AA-Omniscience Hallucination Rate Artificial Analysis Omniscience Hallucination Rate | 60.8 | #127 | Knowledge questions | Factuality | Artificial Analysis model benchmarks |
| HealthBench Professional HealthBench Professional | 59.8 | #15 | Clinician chat tasks | Professional clinical workflows | HealthBench Professional: Evaluating Large Language Models on Real Clinician Chats |
| HLE w/o tools Humanity's Last Exam without tools | 56.3 | #6 | Expert-level questions | Frontier expert level | Introducing GPT-5.4 mini and nano |
| GPQA Diamond (Vals) GPQA Diamond, Vals AI run | 93.4 | #7 | Graduate-level science questions | Expert reasoning | Vals AI GPQA Diamond, Vals AI run leaderboard |
| MMLU-Pro (Vals) MMLU-Pro, Vals AI run | 91.6 | #2 | Academic multiple-choice questions | Broad academic knowledge | Vals AI MMLU-Pro, Vals AI run leaderboard |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| ProgramBench (episode 1) ProgramBench hidden-test pass rate after episode 1 | 83.0 | #1 | 166 golden program-reconstruction tasks | Long-context clean-room software engineering | ProgramBench: Can language models rebuild programs from scratch? |
| SWE-bench Verified Software Engineering Benchmark Verified | 96 | #1 | 500 verified issues | Professional software engineering | SWE-bench: Can Language Models Resolve Real-World GitHub Issues? |
| SWE-bench Pro SWE-bench Pro | 79.2 |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| IMO 2026 International Mathematical Olympiad 2026 | 42 | #1 | 6 proof-based problems | International olympiad mathematics | Claude Opus 5 System Card |
| RiemannBench (no tools) RiemannBench without tools | 60.0 | #1 | 25 private research-level mathematics problems | Research mathematics | Claude Opus 5 System Card |
| RiemannBench (tools) RiemannBench with tools | 79.0 |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| ARC-AGI-1 ARC-AGI-1 Semi-Private Evaluation | 97.50 | #5 | Semi-private ARC-AGI-1 evaluation set | Abstract visual reasoning | ARC Prize leaderboard |
| ARC-AGI-2 Abstraction and Reasoning Corpus for AGI v2 | 90.42 | #8 | Visual pattern completion and abstract reasoning | Expert-level — hardest public reasoning benchmark | ARC-AGI-2: A Harder General Intelligence Benchmark |
| ARC-AGI-3 Abstraction and Reasoning Corpus for AGI v3 | 30.16 |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| GMMLU Global MMLU | 92.5 | #4 | Knowledge questions across 42 languages | Multilingual knowledge | Global MMLU: Understanding and addressing cultural and linguistic biases in multilingual evaluation |
| MILU Multi-task Indic Language Understanding Benchmark | 92.1 | #4 | Knowledge tasks across 11 languages | Multilingual Indic knowledge | MILU: A Multi-task Indic language understanding benchmark |
| INCLUDE INCLUDE | 89.8 |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| DRACO Data Research and Analysis with Complex Operations | 88.6 | #1 | Agentic data research and analysis tasks | Professional data analysis | Claude Opus 5 System Card |
| BrowseComp (10-agent, prerelease) Multi-Agent BrowseComp — 10-agent team prerelease configuration | 93.6 | #1 | BrowseComp web-research tasks | Long-horizon web research | Claude Opus 5 System Card |
| MCP-Atlas claim coverage MCP-Atlas mean claim coverage |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| Chartography (no tools) Chartography without tools | 29.6 | #11 | 100 specialized chart types | Professional chart reasoning | Chartography |
| Chartography (tools) Chartography with image and code tools | 83.0 | #6 | 100 specialized chart types | Professional chart reasoning | Chartography |
| BenchCAD Vision2Code (no tools) BenchCAD Vision2Code voxel IoU without tools | 0.366 |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| Gray Swan IPI (15 attempts) Gray Swan Indirect Prompt Injection, attack success at 15 attempts | 4.6 | #10 | Indirect prompt-injection attacks | Adversarial prompt injection | Gemini 4 Argon evaluation methodology |
| Benchmark | Value | Rank | Tasks | Difficulty | Provenance |
|---|---|---|---|---|---|
| MaxxBench MaxxBench (mutated DeepSWE tasks) | 63.5 | #5 | 89 mutated DeepSWE feature-implementation tasks, two attempts per model | Long-horizon software engineering under changed requirements | Why MaxxBench (unLAP research post) |
| Chess Puzzles Chess Puzzles (Epoch AI) | 43.0 | #33 | 100 programmatically generated chess positions with a single best move | Novel chess positions, single best move | Chess Puzzles (Epoch AI) |
| Mystery Game Puzzles Mystery Game Puzzles (Epoch AI) | 59.0 |
A second, independent score source. Epoch’s own figures in Epoch’s units, one line per model version Epoch evaluated (effort setting as Epoch labels it). They are not part of this site’s capability scores or rank.
| Benchmark | Variant | Score | Epoch date | Evidence |
|---|---|---|---|---|
| GPQA Diamond Accuracy on graduate-level science multiple-choice questions (best score across scorers). Unit: fraction (0–1); higher is better. | default | 0.9293 ± 0.0183 | 2026-08-06 | |
| low | 0.8788 ± 0.0233 | 2026-08-06 | ||
| max | 0.9388 ± 0.0148 | 2026-07-24 | ||
| FrontierMath Tiers 1-3 (v2) Share of private research-level math problems solved, tiers 1-3. Unit: fraction (0–1); higher is better. | max | 0.8561 ± 0.0208 | 2026-07-24 | |
| FrontierMath Tier 4 (v2) Share of private research-level math problems solved, tier 4. Unit: fraction (0–1); higher is better. | max | 0.7317 ± 0.0701 | 2026-07-24 | |
| SimpleQA Verified Short factual questions answered correctly. Unit: fraction (0–1); higher is better. | max | 0.599 ± 0.0155 | 2026-08-10 | |
| OSWorld 2.0 Binary task success on computer-use tasks, as compiled by Epoch from the OSWorld 2.0 leaderboard. Unit: fraction (0–1); higher is better. | high batch tool | 0.2898 | — | |
| low batch tool | 0.2232 | — | ||
| max batch tool | 0.3143 | — | ||
| medium batch tool | 0.2533 | — | ||
| xhigh batch tool | 0.3021 | — |
Source: Epoch AI, CC BY 4.0 (licence), benchmark data read once a day.
People's votes between two anonymous answers on arena.ai, one board per task. LMArena's own ratings and ranks, one line per model version it rates; not part of this site's capability scores or rank.
| Board | Variant | Rating | Votes | Rank | Published | Evidence |
|---|---|---|---|---|---|---|
| Vision Arena People's votes between two answers about the same image, style-controlled. | High | 1,285 ± 6 | 19,229 | 20 | 2026-11-19 |
Source: LMArena, CC BY 4.0 (licence), leaderboard data read once a day.
| #7 |
| 1,865 repository problems |
| Long-horizon professional engineering |
| SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks? |
| VulcanBench v3 VulcanBench v3 | 87.0 | #4 | 23 post-cutoff repository tasks in the v3 report | Professional multi-file software engineering | VulcanBench |
| VulcanBench CII v1 VulcanBench Coding Intelligence Index v1 | 96.4 | #1 | 38 validated post-cutoff repository tasks | Mid-band frontier software engineering | CII v1 frontier results |
| FrontierCode 1.1 Main FrontierCode 1.1 Main | 53.4 | #3 | 100 private Main tasks (150 in Extended) | Frontier coding-agent quality | FrontierCode leaderboard |
| FrontierCode 1.1 Extended FrontierCode 1.1 Extended | 63.6 | #2 | 150 private software-engineering tasks | Frontier coding-agent quality | FrontierCode leaderboard |
| SWE Multilingual SWE Multilingual | 89.5 | #4 | Multilingual software-engineering tasks | Professional software engineering | MiniMax M2.7: Early Echoes of Self-Evolution |
| SWE Multimodal SWE-bench Multimodal | 59.4 | #3 | Multimodal software engineering tasks | Frontier multimodal coding | SWE-bench Multimodal |
| ProgramBench ProgramBench: Can Language Models Rebuild Programs From Scratch? | 93.0 | #1 | 200 program reconstruction tasks | Full-repository software architecture | ProgramBench: Can Language Models Rebuild Programs From Scratch? |
| FrontierSWE v2 FrontierSWE v2 | 52.0 | #10 | 34 ultra-long-horizon engineering and research tasks | Ultra-long-horizon frontier software engineering | FrontierSWE v2 |
| Bug Hunt Bench Bug Hunt Bench | 27.0 | #20 | 105 planted bugs across two production TypeScript repositories | Blind production-repository bug finding and repair | Bug Hunt Bench method, caveats, and definitions |
| AA Coding Index Artificial Analysis Coding Index | 78.0 | #2 | Cross-benchmark coding index | Display-only external reference | Artificial Analysis model leaderboards |
| AA-SciCode Artificial Analysis SciCode | 56.4 | #26 | Scientific coding subproblems | Scientific programming | Artificial Analysis SciCode Benchmark Leaderboard |
| LiveCodeBench (Vals) LiveCodeBench, Vals AI run | 89.0 | #4 | Competitive programming problems (easy, medium, hard) | Frontier coding | Vals AI LiveCodeBench, Vals AI run leaderboard |
| SWE-bench (Vals) SWE-bench, Vals AI run | 97.0 | #1 | Real repository issues by human time bucket | Frontier coding agents | Vals AI SWE-bench, Vals AI run leaderboard |
| DeepSWE DeepSWE | 68.8 | #28 | 113 software engineering tasks across 91 repositories and 5 languages | Long-horizon software engineering | DeepSWE benchmark blog |
| PostTrainBench v1.1 PostTrainBench v1.1 | 35.0 | #7 | Post-training Qwen3 1.7B and 4B, SmolLM3 3B, and Gemma3 4B | Frontier agent evaluation | PostTrainBench v1.1 leaderboard and methodology |
| FrontierCode 2.0 FrontierCode 2.0 | 28.1 | #17 | 132 private software-engineering tasks | Frontier coding-agent quality | FrontierCode 2.0 |
| ScarfBench ScarfBench: Cross-Framework Application Migration in Enterprise Java | 69.8 | #3 | 204 framework migration tasks across 34 enterprise Java applications | Whole-application framework migration | ScarfBench: A Benchmark for Cross-Framework Application Migration in Enterprise Java |
| PostTrainBench PostTrainBench v1.2 | 35.7 | #10 | Post-training four base models on six benchmarks | Autonomous ML engineering | PostTrainBench |
| #1 |
| 25 private research-level mathematics problems |
| Research mathematics |
| Claude Opus 5 System Card |
| ArXivMath Jun. 2026 (no tools) ArXivMath June 2026 without tools | 90.8 | #1 | 49 recent research-mathematics problems | Research mathematics | Claude Opus 5 System Card |
| ArXivMath Jun. 2026 (tools) ArXivMath June 2026 with tools | 91.3 | #1 | 49 recent research-mathematics problems | Research mathematics | Claude Opus 5 System Card |
| FrontierMath v2 (Tiers 1-3) FrontierMath v2 Tiers 1-3 | 85.614 | #17 | 285 private advanced mathematics problems | From olympiad-plus to early research mathematics | FrontierMath v2 benchmark hub |
| FrontierMath v2 (Tier 4) FrontierMath v2 Tier 4 | 73.171 | #17 | 41 private extreme-difficulty mathematics problems | Research-level mathematics requiring hours or days of expert work | FrontierMath Tier 4 v2 leaderboard |
| #8 |
| Interactive game-like tasks with hidden rules |
| Frontier agentic reasoning |
| ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence |
| AA-LCR Artificial Analysis Long Context Reasoning | 79.3 | #55 | Long-context reasoning tasks | Long-context reasoning | Artificial Analysis model benchmarks |
| MLCR-AA Medical Long Context Reasoning (MLCR-AA) | 55.6 | #2 | Long, fragmented medical-record reasoning | Long-context medical reasoning | Medical Long Context Reasoning (MLCR-AA) |
| CritPt Critical Physics Tasks | 29.1 | #13 | Research-level physics questions | Research-level physics reasoning | CritPt Benchmark Leaderboard |
| #1 |
| Cross-lingual understanding |
| Broad multilingual capability |
| Qwen3.6 launch benchmarks |
| 89.1 |
| #1 |
| Production-like multi-server MCP workflows |
| Real-world tool use |
| Claude Opus 5 System Card |
| LAB all-pass (Anthropic harness) Legal Agent Benchmark all-pass rate — Anthropic harness | 23.58 | #1 | 1,235 legal-agent tasks | Professional legal work | Claude Opus 5 System Card |
| LAB criterion-pass (Anthropic harness) Legal Agent Benchmark mean criterion-pass rate — Anthropic harness | 93.74 | #1 | 1,235 legal-agent tasks | Professional legal work | Claude Opus 5 System Card |
| LAB all-pass (Harvey held-out) Legal Agent Benchmark all-pass rate — Harvey held-out set | 11.7 | #1 | Harvey-held-out legal-agent tasks | Professional legal work | Claude Opus 5 System Card |
| LAB criterion-pass (Harvey held-out) Legal Agent Benchmark mean criterion-pass rate — Harvey held-out set | 94.1 | #1 | Harvey-held-out legal-agent tasks | Professional legal work | Claude Opus 5 System Card |
| Toolathlon Verified Pass@3 Toolathlon Verified Pass@3 | 87.0 | #1 | 108 verified real-world tool-use tasks | Long-horizon application tool use | Claude Opus 5 System Card |
| Toolathlon Verified Pass³ Toolathlon Verified Pass cubed | 73.1 | #1 | 108 verified real-world tool-use tasks | Long-horizon application tool use | Claude Opus 5 System Card |
| Toolathlon Verified avg. turns Toolathlon Verified average assistant turns | 23.5 | #4 | 108 verified real-world tool-use tasks | Long-horizon application tool use | Claude Opus 5 System Card |
| Terminal-Bench 3.0 Terminal-Bench 3.0 | 42.7 | #1 | 74 professional computer-work tasks across 7 domains | Frontier autonomous knowledge work | Terminal-Bench 3.0 |
| AA Briefcase Artificial Analysis Briefcase | 1720 | #5 | Professional knowledge-work tasks | Professional work | Artificial Analysis Briefcase Benchmark Leaderboard |
| AA AutomationBench Artificial Analysis AutomationBench | 56.6 | #32 | Business-process automation tasks | Agentic automation | Artificial Analysis AutomationBench Benchmark Leaderboard |
| AA EnterpriseOps-Gym Artificial Analysis EnterpriseOps-Gym | 47.5 | #7 | Enterprise operations workflows | Enterprise agent operations | Artificial Analysis EnterpriseOps-Gym Benchmark Leaderboard |
| AA Harvey LAB v1.0 Artificial Analysis Harvey LAB-AA v1.0 | 93.5 | #3 | Legal agent tasks | Professional legal work | Artificial Analysis Harvey LAB-AA Benchmark Leaderboard |
| AA Tau3 Banking Artificial Analysis Tau3-Banking | 42.1 | #11 | Banking tool-use workflows | Agentic banking workflows | Artificial Analysis Tau3-Banking Benchmark Leaderboard |
| BrowseComp BrowseComp | 90.8 | #8 | Research questions requiring browsing | Hard web research | BrowseComp |
| HLE w/ tools Humanity's Last Exam with tools | 64.7 | #3 | Expert questions with tool use | Frontier tool-augmented reasoning | DeepSeek-V4 Technical Report |
| GDPval-AA GDPval-AA | 1862 | #1 | Agentic real-world work tasks | Professional agentic workflows | DeepSeek-V4 Technical Report |
| GDPval-AA GDPval-AA normalized | 61.1 | #7 | Economically valuable tasks | Professional agentic workflows | Artificial Analysis model benchmarks |
| AA Agentic Index Artificial Analysis Agentic Index | 56.2 | #2 | Cross-benchmark agentic index | Display-only external reference | Artificial Analysis model leaderboards |
| AA Terminal-Bench 4.0 Artificial Analysis Terminal-Bench v4.0 | 49.0 | #17 | Terminal-based agent tasks | Agentic software engineering | Artificial Analysis Terminal-Bench v4.0 Benchmark Leaderboard |
| GDP.pdf Artificial Analysis GDP.pdf | 21.6 | #28 | Professional document-production tasks | Professional knowledge work | GDP.pdf Benchmark Leaderboard |
| AA-AnalystAgent Artificial Analysis AnalystAgent | 53.8 | #6 | Spreadsheet and document analysis questions | Business and data analysis | AA-AnalystAgent Benchmark Leaderboard |
| OSWorld 2.0 OSWorld 2.0 | 70.6 | #3 | 108 long-horizon computer-use workflows | Long-horizon professional workflows | OSWorld2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks |
| MCP Atlas MCP Atlas | 85.8 | #3 | Tool-integrated agent tasks | Advanced tool use | Introducing GPT-5.4 mini and nano |
| Toolathlon-Verified Toolathlon-Verified | 80.6 | #1 | Verified multi-tool workflows | Advanced tool use | Kimi K3: Open Frontier Intelligence |
| AutomationBench AutomationBench | 26.0 | #33 | 600 public automation tasks | Long-horizon automation | Kimi K3: Open Frontier Intelligence |
| DeepSearchQA DeepSearchQA | 95.0 | #2 | Agentic browsing and list-answer questions | Agentic web research | Muse Spark Eval Methodology |
| Terminal-Bench 2.1 (Vals) Terminal-Bench 2.1, Vals AI run | 84.6 | #5 | Difficult terminal tasks | Frontier agentic | Vals AI Terminal-Bench 2.1, Vals AI run leaderboard |
| ApprenticeBench ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable job | 36 | #3 | 100 vendor bills processed in sequence inside a simulated construction company | Long-horizon computer use with offline and online continual learning | ApprenticeBench: a step change in AI's job readiness |
| MazeBench MazeBench: visual spatial reasoning in a 3D open world | 12 | #13 | 100 hidden gems across 256 rooms in one open world | Long-horizon spatial planning, with single puzzles that can take over one hundred moves | MazeBench Results |
| Terminal-Bench 4.1 Terminal-Bench 4.1 | 53.30 | #14 | 66 professional computer-work tasks | Frontier autonomous knowledge work | Terminal-Bench 4.1 |
| OSWorld 2.1 OSWorld 2.1 (all 108 tasks) | 74.0 | #8 | 108 long-horizon computer-use workflows | Long-horizon professional workflows | OSWorld 2.1 release (xlang-ai/OSWorld-V2, tag osworld-v2.1) |
| DeepsecBench-AA Artificial Analysis DeepsecBench-AA | 25.9 | #10 | Vulnerability discovery in real codebases | Security code review | DeepsecBench-AA Benchmark Leaderboard |
| #5 |
| 1,000-file Vision2Code subset |
| Programmatic CAD generation |
| BenchCAD: A comprehensive, industry-standard benchmark for programmatic CAD |
| BenchCAD Vision2Code (tools) BenchCAD Vision2Code voxel IoU with tools | 0.821 | #6 | 1,000-file Vision2Code subset | Programmatic CAD generation | BenchCAD: A comprehensive, industry-standard benchmark for programmatic CAD |
| GDP.pdf (no tools) GDP.pdf mean criteria pass rate without tools | 83.4 | #2 | 100 professional document prompts | Professional document reasoning | GDP.pdf |
| GDP.pdf (tools) GDP.pdf mean criteria pass rate with tools | 85.5 | #1 | 100 professional document prompts | Professional document reasoning | GDP.pdf |
| OfficeQA OfficeQA | 78.1 | #4 | Historical Treasury Bulletin questions | Professional document reasoning | Claude Opus 5 System Card |
| AA-MMMU-Pro Artificial Analysis MMMU-Pro | 84.7 | #7 | Multimodal academic reasoning | Frontier multimodal | Artificial Analysis MMMU-Pro Benchmark Leaderboard |
| Design Arena Website Design Arena Website Elo | 1299 | #18 | Website generation comparisons | Design and website generation | OpenRouter Grok 4.3 benchmarks |
| OfficeQA Pro OfficeQA Pro | 66.9 | #5 | Document and spreadsheet tasks | Enterprise grounded reasoning | OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoning |
| #12 |
| 100 single-best-move positions from an undisclosed game |
| Unfamiliar game positions, single best move |
| Mystery Game Puzzles (Epoch AI) |
| Furniture Assembly Furniture Assembly Benchmark (Epoch AI) | 60.8 | #12 | 60 photos of three IKEA builds, some with deliberate assembly mistakes | Spatial reasoning from technical diagrams to real photos | Furniture Assembly Benchmark (Epoch AI) |
| EBR-bench Earthborne Rangers (EBR-bench) | 45.7 | #12 | Ten playthroughs of an Earthborne Rangers campaign segment per sample (eight learning, two scored) | Learning an unfamiliar game over multi-hour playthroughs | Earthborne Rangers (EBR-bench), Epoch AI |
| MirrorCode MirrorCode (Epoch AI and METR) | 87.8 | #4 | 30 reimplementation tasks (15 programs in two languages), three attempts each | Days-long software engineering without source code | MirrorCode (Epoch AI) |
| DAYJOB DAYJOB | 11.7 | #7 | 130 long-horizon tasks: 80 finance and 50 healthcare | Long-horizon professional work | BenchLM |
| DAYJOB Healthcare DAYJOB Healthcare | 9.2 | #7 | 50 long-horizon healthcare tasks across five subfields | Long-horizon clinical and healthcare operations work | BenchLM |
| Code2Video Code2Video | 1543 | #6 | Creative briefs rendered as motion-graphics video | Code-driven motion design | BenchLM |
| Joke Arena Joke Arena | 1511 | #8 | 100 joke prompts in 10 categories, rated by human votes | Humour writing | BenchLM |
| Pelican Arena Pelican Arena | 1560 | #14 | One prompt: an animated SVG of a pelican riding a bicycle, rated by human votes | Vector drawing and animation in code | BenchLM |
| ASCII Arena ASCII Arena | 1483 | #9 | ASCII drawings of subjects from Google's Quick, Draw! dataset, rated by human votes | Text-character drawing | BenchLM |
| Game Arena Go Game Arena Go | 1564 | #6 | Head-to-head Go games on a 13 x 13 board | Strategic planning and rule-following | BenchLM |
| BabaIsBench BabaIsBench (Baba Is You intro levels, pass@1) | 100 | #1 | 8 intro levels of Baba Is You, three attempts per level (Stage 0); 15 further levels run once (Stage 1) | Rule-rewriting puzzles that need planning over many moves | Baba Is Solved by Fable 5 and GPT-5.6 Sol, but at what cost? (Quesma) |
| NYT Connections NYT Connections LLM Benchmark (extended version) | 94.3 | #10 | 940 NYT Connections puzzles, each with up to four added trick words | Wordplay, word knowledge and decoys that fit almost-but-not-quite | NYT Connections LLM Benchmark (Lech Mazur, GitHub) |
| FrontierPhysics End-to-End Physics Research Benchmark for AI Agents | 9.5 | #13 | 56 research projects from 42 physicists across nine fields, three trials each (168 rollouts per setup) | Open-ended research projects that took their authors about 270 days on average | FrontierPhysics: Evaluating Agents for End-to-End Physics Research (arXiv, November 4, 2026) |