Claude Opus · 发布历史
Anthropic Claude Opus 系列已收录 6 次发布和 115 个测试版本。
Claude Opus 旗舰产品线;型号及前代关系依据 Anthropic 官方发布材料。
全部版本
- Claude Opus 5.541 项测试
- Claude Opus 546 项测试
- Claude Opus 4.835 项测试
- Claude Opus 4.735 项测试
- Claude Opus 4.635 项测试
- Claude Opus 4.531 项测试
已收录测试 · 115
- Terminal-Bench · 4.0Claude Opus 5.5
- FrontierCode · v1.1 Main setClaude Opus 5.5、Claude Opus 5
- CursorBench · 4.0Claude Opus 5.5
- GDPval-AA · v2.1Claude Opus 5.5
- AutomationBench · Zapier 私有 early-access set;版本未注明Claude Opus 5.5、Claude Opus 5
- Humanity's Last Exam · 2,500-question benchmarkClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- Terminal-Bench-Science · 0.1Claude Opus 5.5
- OSWorld · 2.1 (September 10, 2026 task/assets release)Claude Opus 5.5
- Chartography · 100 tasksClaude Opus 5.5、Claude Opus 5
- OSWorld · System Card 所称 2.0;2026-09-10 任务与资源版本、108 题、更新 harnessClaude Opus 5.5、Claude Opus 5
- OSWorld · 2.1, 82-task offline no-internet subset, partial scoreClaude Opus 5.5
- SWE-bench Pro · ProClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.5
- DeepSWE · v1.1Claude Opus 5.5、Claude Opus 5
- FrontierSWE · v2Claude Opus 5.5
- DRACO · 100-task Perplexity deep-research benchmarkClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- WANDR · 500-task wide research benchmarkClaude Opus 5.5
- ProgramBench · 166 个保留的 golden tasks;架构与时限按协议区分Claude Opus 5.5、Claude Opus 5
- BenchCAD Vision2Code · 1,000-file random subset of 17,900 filesClaude Opus 5.5、Claude Opus 5
- OfficeQA · Full setClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- Legal Agent Benchmark · Opus 5 官方发布表所列 held-out setClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8
- AA-Briefcase · v1.1Claude Opus 5.5
- Toolathlon-Verified · June 2026 finalized releaseClaude Opus 5.5
- HealthBench · 48,000+ rubric items across 5,000 multi-turn conversationsClaude Opus 5.5、Claude Opus 5
- HealthBench Professional · 525 physician-authored conversationsClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8
- Global MMLU · 42 languages including English · overall averageClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall averageClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- SpatialBench Verified · VerifiedClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8
- Morphology-to-molecule matching · Four multi-channel human-liver in-vitro cell imaging sets matched to shuffled molecule-dose perturbationsClaude Opus 5.5
- Medicinal Chemistry ADME · 504 drug-discovery questions about measured ADME propertiesClaude Opus 5.5
- Protein Design Sequence Generation · Protein sequence generation conditioned on design constraintsClaude Opus 5.5、Claude Opus 5
- De novo protein binder design · System Card evaluation; 15 targets; 30 miniprotein binders per target; Cas9 excludedClaude Opus 5.5
- Biomedical Image Analysis · System-card evaluation across public biomedical-image competition tasks; included competition set unspecifiedClaude Opus 5.5
- Protocol Troubleshooting · System-card protocol troubleshooting variant; task count unspecifiedClaude Opus 5.5、Claude Opus 5
- RoboDojo Sim · 版本/任务配置未注明Claude Opus 5.5
- SWE-bench Multilingual · 300 problems, 9 programming languagesClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- SWE-bench Multimodal · MultimodalClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- FrontierCode · v1.1 Extended setClaude Opus 5.5、Claude Opus 5
- OfficeQA · Pro 133-question subsetClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- SingleCellBench · LatchBio Bioinformatics single-cell RNA sequencing analysis evaluationClaude Opus 5.5、Claude Opus 5、Claude Opus 4.8
- Protein Design Library Ranking · Prioritization of protein designs for experimental testingClaude Opus 5.5
- Protocol Understanding · V2; Benchling-created evaluation using real laboratory protocols and applied questions; task count unspecifiedClaude Opus 5.5、Claude Opus 5
- Frontier-Bench · v0.1Claude Opus 5
- GDPval-AA · v2Claude Opus 5
- ARC-AGI-3 · 版本/子集未注明Claude Opus 5
- BrowseComp · 1,266 题;单代理与多代理按协议区分Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- OSWorld · 2.0;Opus 5 发布表,未注明评测子集/协议Claude Opus 5
- BioMysteryBench · 官方发布表列 Hard 与 human-solved 结果;细分版本未注明Claude Opus 5、Claude Opus 4.6
- SWE-bench Verified · Verified subset (500 problems)Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- International Mathematical Olympiad · 2026 contest; six proof problemsClaude Opus 5
- RiemannBench · 25 private research-mathematics problems; corrected reference answers and grading; four attempts per problemClaude Opus 5
- ArXivMath · June 2026 release (49 problems)Claude Opus 5
- DeepSearchQA · 900 prompts;单代理与多代理按协议区分Claude Opus 5、Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- GDP.pdf · 100 professional PDFs and promptsClaude Opus 5
- MCP-Atlas · 所选材料未明确子集Claude Opus 5、Claude Opus 4.6、Claude Opus 4.5
- Legal Agent Benchmark · Public full setClaude Opus 5
- AA-Briefcase · Not stated in selected sourceClaude Opus 5
- ARC-AGI-1 · ARC Prize Foundation semi-private 2026 evaluation setClaude Opus 5
- ARC-AGI-2 · ARC Prize Foundation semi-private 2026 evaluation setClaude Opus 5
- ARC-AGI-3 · ARC Prize Foundation semi-private 2026 evaluation setClaude Opus 5
- INCLUDE · 44 languagesClaude Opus 5、Claude Opus 4.8、Claude Opus 4.7
- ProteinGym Hard · System-card Hard subset; exact task count not statedClaude Opus 5
- Organic Chemistry · V2; system-card evaluation with expert-recommended harder problemsClaude Opus 5
- Terminal-Bench · 2.1Claude Opus 4.8
- OSWorld-Verified · Verified benchmark; evaluation protocol noted per sourceClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6
- GDPval-AA · Artificial Analysis leaderboardClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- Finance Agent · v2Claude Opus 4.8、Claude Opus 4.7
- GPQA Diamond · 198-question Diamond subsetClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- USA Mathematical Olympiad · 2026Claude Opus 4.8、Claude Opus 4.7
- ArXivMath · March-April 2026 releases; 71 problemsClaude Opus 4.8
- GraphWalks · 256K-context subset of the 256K-1M evaluation; item count unspecified; BFS F1Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6
- GraphWalks · 256K-context subset of the 256K-1M evaluation; item count unspecified; parentsClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6
- ChartQAPro · ChartQAPro; set size and release version not stated in the System CardClaude Opus 4.8
- ChartMuseum · 1,162 test-split questionsClaude Opus 4.8
- LAB-Bench FigQA · FutureHouse FigQA; prompting and tools are recorded per resultClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- CharXiv Reasoning · Validation splitClaude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6
- ScreenSpot-Pro · 1,581 expert-annotated tasks across 23 professional applications and 3 operating systemsClaude Opus 4.8、Claude Opus 4.7
- MCP-Atlas · Scale AI 公共排行榜;2026-04 刷新 harnessClaude Opus 4.8、Claude Opus 4.7
- Vending-Bench · 2Claude Opus 4.8、Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- Toolathlon · June 2026 release used in selected sourceClaude Opus 4.8
- AutomationBench · Zapier private set; release/version not statedClaude Opus 4.8
- Terminal-Bench · 2.0;代理框架按协议区分Claude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- Finance Agent · Vals AI SEC-filing research benchmarkClaude Opus 4.7、Claude Opus 4.5
- MMMLU · 57 subjects and 14 non-English languagesClaude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- GraphWalks · 200 problems, 256k to 1M context, BFS F1Claude Opus 4.7
- OpenAI MRCR · v2, 256K context, 8 needlesClaude Opus 4.7
- OpenAI MRCR · v2, 1M context, 8 needlesClaude Opus 4.7
- ARC-AGI-1 · ARC Prize Foundation private datasetClaude Opus 4.7、Claude Opus 4.5
- ARC-AGI-2 · ARC Prize Foundation private datasetClaude Opus 4.7、Claude Opus 4.6、Claude Opus 4.5
- GraphWalks · parents;256K–1M 上下文;题数未注明Claude Opus 4.7、Claude Opus 4.6
- Structural Biology · Opus 4.7 announcement figure; dataset, version and question format unspecifiedClaude Opus 4.7
- τ²-bench · Retail sectionClaude Opus 4.6、Claude Opus 4.5
- τ²-bench · Telecom sectionClaude Opus 4.6、Claude Opus 4.5
- Finance Agent · Opus 4.6 来源未注明版本;发布于 Vals AI v2 (2026-05) 之前Claude Opus 4.6
- ARC-AGI-2 · Verified;题集组成未在发布表中说明Claude Opus 4.6、Claude Opus 4.5
- MMMU-Pro · Standard and Vision formatsClaude Opus 4.6、Claude Opus 4.5
- OpenRCA · 335 enterprise software failure casesClaude Opus 4.6、Claude Opus 4.5
- AIME · 2025 editionClaude Opus 4.6、Claude Opus 4.5
- Real-World Finance · Anthropic internal evaluation; about 50 analyst-workflow tasks; version unspecifiedClaude Opus 4.6
- OpenAI MRCR v2 · v2 fix dated 2025-12-05; 8 needles; prompts in (128K, 256K] tokens; Mean Match RatioClaude Opus 4.6
- OpenAI MRCR v2 · v2 fix dated 2025-12-05; 8 needles; prompts in (524K, 1024K] tokens; full variant may exceed the public API context limitClaude Opus 4.6
- GraphWalks · BFS;256K–1M 上下文;题数未注明Claude Opus 4.6
- WebArena · Official WebArena grader, modified fuzzy_match subgraderClaude Opus 4.6
- BioPipelineBench · Anthropic internal unpublished computational-biology evaluation; version and task count unspecifiedClaude Opus 4.6
- Structural Biology · Anthropic internal unpublished life-science evaluation; format recorded per resultClaude Opus 4.6
- Organic Chemistry · Anthropic internal unpublished life-science evaluation; version unspecifiedClaude Opus 4.6
- Phylogenetics · Anthropic internal unpublished life-science evaluation; version unspecifiedClaude Opus 4.6
- OSWorld · 版本未注明;Claude Opus 4.5 发布材料所列Claude Opus 4.5
- MMMU · Validation splitClaude Opus 4.5
- τ²-bench · Anthropic corrected Airline variant; task and grading corrections submitted to evaluation authorsClaude Opus 4.5
- tau2-bench · Airline categoryClaude Opus 4.5
- SpreadsheetBench · Full 912-problem setClaude Opus 4.5
- WebArena · Official WebArena grader; exact task release/subset not stated in cardClaude Opus 4.5
- Simple-QA-Verified · Anthropic factuality evaluation; Claude Sonnet 4.5 grader; exact dataset revision not statedClaude Opus 4.5
- AA-Omniscience · Anthropic factuality evaluation on 42 economically relevant topics across six domains; Claude Sonnet 4.5 grader; dataset revision not statedClaude Opus 4.5
- Aider Polyglot · Diff or diff-fenced edit formatClaude Opus 4.5