Claude Sonnet 4.6 · Humanity's Last Exam(2,500-question benchmark),Adaptive max, without tools配置,Accuracy 33.2%(具体成绩公布时间未确认)。见《Claude Sonnet 4.6 System Card》§2.1 Table 2.1.A, PDF p.15, HLE rows。 查看成绩条件
Claude Sonnet 4.6 · Humanity's Last Exam(2,500-question benchmark),Adaptive thinking, max effort配置,Accuracy 49%(具体成绩公布时间未确认)。见《Claude Sonnet 4.6 System Card》§2.1 Table 2.1.A, PDF p.15, HLE rows。 查看成绩条件
HLE | No tools | Claude Sonnet 4.6: 33.2%. With tools | Claude Sonnet 4.6: 49.0%.
PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本
Claude Sonnet 4.6 · Humanity's Last Exam(2,500-question benchmark),Adaptive max, without tools配置,Accuracy 33.2%(具体成绩公布时间未确认);Claude Sonnet 4.6 · Humanity's Last Exam(2,500-question benchmark),Adaptive thinking, max effort配置,Accuracy 49%(具体成绩公布时间未确认)。官方出处 Claude Sonnet 4.6 System Card(附件,PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10)§2.1 Table 2.1.A, PDF p.15, HLE rows,https://www.anthropic.com/claude-sonnet-4-6-system-card#page=15。Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-4-6/ev46-card-hle/
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
未知项
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
工具
允许工具
推理预算
max
框架
未记录
协议
Humanity's Last Exam, 2,500 questions; web search, web fetch, code execution and programmatic tool calling.
条件
Context compaction at 50K tokens up to 3M total tokens, adaptive thinking, max effort; domain blocklist and transcript review used for contamination checks.
Claude Sonnet 4.6 · Humanity's Last Exam (2,500-question benchmark) · Attachment evidence (ev46-card-hle)
Official source for Humanity's Last Exam · 2,500-question benchmark in Claude Sonnet 4.6. See “Claude Sonnet 4.6 System Card” at §2.1 Table 2.1.A, PDF p.15, HLE rows. Evidence ev46-card-hle.
Claude Sonnet 4.6 · Humanity's Last Exam (2,500-question benchmark), configuration Adaptive max, without tools, Accuracy 33.2% (exact score publication date unconfirmed). See “Claude Sonnet 4.6 System Card” at §2.1 Table 2.1.A, PDF p.15, HLE rows. View score conditions
Claude Sonnet 4.6 · Humanity's Last Exam (2,500-question benchmark), configuration Adaptive thinking, max effort, Accuracy 49% (exact score publication date unconfirmed). See “Claude Sonnet 4.6 System Card” at §2.1 Table 2.1.A, PDF p.15, HLE rows. View score conditions
HLE | No tools | Claude Sonnet 4.6: 33.2%. With tools | Claude Sonnet 4.6: 49.0%.
PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10
Version published
Unconfirmed
Disclosure date
Unconfirmed · Launch or later disclosure unconfirmed
Collected
Historical basis
No reliable historical version evidence available
Citation text
Claude Sonnet 4.6 · Humanity's Last Exam (2,500-question benchmark), configuration Adaptive max, without tools, Accuracy 33.2% (exact score publication date unconfirmed); Claude Sonnet 4.6 · Humanity's Last Exam (2,500-question benchmark), configuration Adaptive thinking, max effort, Accuracy 49% (exact score publication date unconfirmed). Official source: Claude Sonnet 4.6 System Card (Attachment, PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10), §2.1 Table 2.1.A, PDF p.15, HLE rows, https://www.anthropic.com/claude-sonnet-4-6-system-card#page=15. Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-4-6/ev46-card-hle/?lang=en
Related scores and conditions
Claude Sonnet 4.6 · Humanity's Last Exam scores and configurations
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
Unknowns
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
Humanity's Last Exam, 2,500 questions; web search, web fetch, code execution and programmatic tool calling.
Conditions
Context compaction at 50K tokens up to 3M total tokens, adaptive thinking, max effort; domain blocklist and transcript review used for contamination checks.