Home Model series Claude Sonnet Claude Sonnet 5.5 · Official benchmark disclosures Claude Sonnet 5.5 · Official benchmark disclosures Claude Sonnet 5.5 has official records for 34 benchmarks. All listed materials have been reviewed.
Model released 2026-09-28 · Claude Sonnet
Official materials and review scope 列明主文、系统卡能力与风险章节、图表和附录已完成模型来源核对;首发历史仍需独立证据。 客户私有测试仅保留官方材料内的引述披露,不参与精确排名。 Checked 2026-11-10 .
Available materials are insufficient to reconstruct the launch state. 已取得首发当天官方主文回放,但尚未取得并核读全部直接附件的可靠历史正文及外链图像历史字节,无法完整还原首发材料集合;成功回放与失败材料分别保留,不用当前附件替代首发。
Claude Sonnet 5.5 · Main text · Current page collected 2026-11-10; page displays September 28, 2026 publication datePublished Unconfirmed · Collected 2026-11-10 · Reviewed 主文全文、汇总表、能力与客户评测段落、脚注均已检查;所有跨模型汇总表只录入 Sonnet 5.5 主体列。
Claude Sonnet 5.5 System Card · Attachment · PDF cover dated 2026-09-28; 148 pages; collected 2026-11-10Published Unconfirmed · Collected 2026-11-10 · Reviewed 系统卡能力、风险、安全、对齐、模型福利与附录已核读,能力图表已逐页视觉核查。只录当前主体的精确印值;材料完整不表示每项均有可排名成绩。部分缩放图没有精确分数、耗时或成本标签,不从坐标估读。HLE 的 thinking mode 与 effort 分别记录。
Configurations Adaptive thinking, max effort · Adaptive thinking enabled; max effort. maxUnknowns Sampling settings may vary by benchmark; see each result.conditions.
Xhigh effort · Xhigh reasoning effort. xhighUnknowns Sampling settings not stated unless recorded per result.
High effort · High reasoning effort. highUnknowns Sampling settings not stated unless recorded per result.
Low effort · Low reasoning effort. lowUnknowns Sampling settings not stated unless recorded per result.
Medium effort · Medium reasoning effort. mediumUnknowns Sampling settings not stated unless recorded per result.
Max effort · Max reasoning effort. maxUnknowns Adaptive thinking status is benchmark-specific.
HLE · Auto thinking · Max effort · 有工具 · Thinking auto;图中标为 Max effort;web search、web fetch、programmatic tool calling 和 code execution。 maxUnknowns 采样参数未说明。
Adaptive max, no tools · Adaptive thinking and max effort; no tools. maxAdaptive max, with tools · Adaptive thinking and max effort; tools listed in result.conditions. maxTerminal-Bench 4.0 max effort with safeguards · Claude Code --bare mode, max thinking effort, cyber safeguards enabled. maxUnknowns Default sampling settings not stated.
Adaptive max with biology safeguards disabled · Adaptive thinking at max effort; API evaluation with biology safeguards turned off. maxUnknowns Sampling settings follow each benchmark's report.
HLE · Auto thinking · Max effort · 无工具 · Thinking auto;Max effort;无工具。 maxUnknowns 采样参数未说明。
Direct predecessor and later releases Direct predecessor Claude Sonnet 5 . Official relationship source
Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Claude Sonnet 5. 14 continued · 20 added · 18 not found · 0 pending review Disclosure changes between Claude Sonnet 5 and Claude Sonnet 5.5 Benchmark Reviewed status Claude Sonnet 5 Claude Sonnet 5.5 Terminal-Bench · 4.0 Added — Current record FrontierCode · v1.1 Main set Added — Current record CursorBench · 4.0 Added — Current record GDPval-AA · v2.1 Added — Current record AA-Briefcase · v1.1 Added — Current record Humanity's Last Exam · 2,500-question benchmark ContinuedRecorded evaluation conditions changed
Predecessor record Current record OSWorld · 2.1 (September 10, 2026 task/assets release) Added — Current record Chartography · 100 tasks Added — Current record SWE-bench Pro · Pro ContinuedSource location attachment、main → attachment
Recorded evaluation conditions changed
Predecessor record Current record DeepSWE · v1.1 Added — Current record FrontierCode · v1.1 Extended set Added — Current record Terminal-Bench-Science · 0.1 Added — Current record FrontierSWE · v2 Added — Current record ArXivMath · August 2026 release (57 problems) Added — Current record ProgramBench · 166 个保留的 golden tasks;架构与时限按协议区分 ContinuedRecorded evaluation conditions changed
Predecessor record Current record DRACO · 100-task Perplexity deep-research benchmark Added — Current record WANDR · 500-task wide research benchmark Added — Current record BenchCAD Vision2Code · 1,000-file random subset of 17,900 files ContinuedRecorded evaluation conditions changed
Predecessor record Current record OfficeQA · Full set ContinuedRecorded evaluation conditions changed
Predecessor record Current record Toolathlon-Verified · June 2026 finalized release Added — Current record AutomationBench · Release 1.0.6, Zapier private evaluation set Added — Current record HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations ContinuedRecorded evaluation conditions changed
Predecessor record Current record HealthBench Professional · 525 physician-authored conversations ContinuedRecorded evaluation conditions changed
Predecessor record Current record Global MMLU · 42 languages including English · overall average ContinuedRecorded evaluation conditions changed
Predecessor record Current record Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall average ContinuedRecorded evaluation conditions changed
Predecessor record Current record SpatialBench Verified · Verified ContinuedRecorded evaluation conditions changed
Predecessor record Current record OfficeQA · Pro 133-question subset ContinuedRecorded evaluation conditions changed
Predecessor record Current record SWE-bench Multilingual · 300 problems, 9 programming languages ContinuedRecorded evaluation conditions changed
Predecessor record Current record SWE-bench Multimodal · Multimodal ContinuedRecorded evaluation conditions changed
Predecessor record Current record PhysicianBench · 100 physician tasks in an electronic health record; pass rate across attempts Added — Current record SingleCellBench · LatchBio Bioinformatics single-cell RNA sequencing analysis evaluation ContinuedRecorded evaluation conditions changed
Predecessor record Current record Morphology-to-molecule matching · Four multi-channel human-liver in-vitro cell imaging sets matched to shuffled molecule-dose perturbations Added — Current record Protein Design Sequence Generation · Protein sequence generation conditioned on design constraints Added — Current record Protein Design Library Ranking · Prioritization of protein designs for experimental testing Added — Current record Terminal-Bench · 2.1 Not found in this release Predecessor record — OSWorld-Verified · Verified benchmark; evaluation protocol noted per source Not found in this release Predecessor record — GDPval-AA · v2 Not found in this release Predecessor record — SWE-bench Verified · Verified subset (500 problems) Not found in this release Predecessor record — FrontierCode · v1 Not found in this release Predecessor record — CursorBench · Not stated in selected source Not found in this release Predecessor record — USA Mathematical Olympiad · 2026 Not found in this release Predecessor record — ArXivMath · April-May 2026 releases (81 problems) Not found in this release Predecessor record — BrowseComp · 1,266 题;单代理与多代理按协议区分 Not found in this release Predecessor record — GDP.pdf · 100 professional PDFs and prompts Not found in this release Predecessor record — ChartMuseum · 1,162 test-split questions Not found in this release Predecessor record — CharXiv Reasoning · Validation split Not found in this release Predecessor record — Legal Agent Benchmark · Public full set Not found in this release Predecessor record — Toolathlon · June 2026 release used in selected source Not found in this release Predecessor record — AA-Briefcase · Not stated in selected source Not found in this release Predecessor record — INCLUDE · 44 languages Not found in this release Predecessor record — Real-World Finance · v2, 294 tasks Not found in this release Predecessor record — ProteinGym Hard · System-card Hard subset; exact task count not stated Not found in this release Predecessor record —
Benchmark records · 34 Terminal-Bench · 4.0 1 score record for this release, including 1 exact value (may include historical records) FrontierCode · v1.1 Main set 2 score records for this release, including 2 exact values (may include historical records) CursorBench · 4.0 4 score records for this release, including 4 exact values (may include historical records) GDPval-AA · v2.1 2 score records for this release, including 2 exact values (may include historical records) AA-Briefcase · v1.1 2 score records for this release, including 2 exact values (may include historical records) Humanity's Last Exam · 2,500-question benchmark 10 score records for this release, including 10 exact values (may include historical records) OSWorld · 2.1 (September 10, 2026 task/assets release) 2 score records for this release, including 2 exact values (may include historical records) Chartography · 100 tasks 2 score records for this release, including 2 exact values (may include historical records) SWE-bench Pro · Pro 1 score record for this release, including 1 exact value (may include historical records) DeepSWE · v1.1 1 score record for this release, including 1 exact value (may include historical records) FrontierCode · v1.1 Extended set 2 score records for this release, including 2 exact values (may include historical records) Terminal-Bench-Science · 0.1 1 score record for this release, including 1 exact value (may include historical records) FrontierSWE · v2 1 score record for this release, including 1 exact value (may include historical records) ArXivMath · August 2026 release (57 problems) 2 score records for this release, including 2 exact values (may include historical records) ProgramBench · 166 个保留的 golden tasks;架构与时限按协议区分 1 score record for this release, including 1 exact value (may include historical records) DRACO · 100-task Perplexity deep-research benchmark 5 score records for this release, including 5 exact values (may include historical records) WANDR · 500-task wide research benchmark 5 score records for this release, including 5 exact values (may include historical records) BenchCAD Vision2Code · 1,000-file random subset of 17,900 files 2 score records for this release, including 2 exact values (may include historical records) OfficeQA · Full set 1 score record for this release, including 1 exact value (may include historical records) Toolathlon-Verified · June 2026 finalized release 4 score records for this release, including 4 exact values (may include historical records) AutomationBench · Release 1.0.6, Zapier private evaluation set 1 score record for this release, including 1 exact value (may include historical records) HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations 5 score records for this release, including 5 exact values (may include historical records) HealthBench Professional · 525 physician-authored conversations 6 score records for this release, including 6 exact values (may include historical records) Global MMLU · 42 languages including English · overall average 1 score record for this release, including 1 exact value (may include historical records) Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall average 1 score record for this release, including 1 exact value (may include historical records) SpatialBench Verified · Verified 1 score record for this release, including 1 exact value (may include historical records) OfficeQA · Pro 133-question subset 1 score record for this release, including 1 exact value (may include historical records) SWE-bench Multilingual · 300 problems, 9 programming languages 1 score record for this release, including 1 exact value (may include historical records) SWE-bench Multimodal · Multimodal 1 score record for this release, including 1 exact value (may include historical records) PhysicianBench · 100 physician tasks in an electronic health record; pass rate across attempts 5 score records for this release, including 5 exact values (may include historical records) SingleCellBench · LatchBio Bioinformatics single-cell RNA sequencing analysis evaluation 1 score record for this release, including 1 exact value (may include historical records) Morphology-to-molecule matching · Four multi-channel human-liver in-vitro cell imaging sets matched to shuffled molecule-dose perturbations 1 score record for this release, including 1 exact value (may include historical records) Protein Design Sequence Generation · Protein sequence generation conditioned on design constraints 1 score record for this release, including 1 exact value (may include historical records) Protein Design Library Ranking · Prioritization of protein designs for experimental testing 1 score record for this release, including 1 exact value (may include historical records)