Claude Sonnet 4.6 · 官方评测披露

Claude Sonnet 4.6 当前收录 31 项测试的官方记录。已完成列明材料的核查。

模型发布 · Claude Sonnet

官方材料与核查范围

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。 核对时间 。

当前资料不足以还原首发。 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。

  • Introducing Claude Sonnet 4.6 · 主文 · Current page collected 2026-11-10; datePublished 2026-02-17; dateModified 2026-10-14T23:56:52Z
    公布 未确认 · 采集 · 已核查 主文全文及脚注已检查;摘要表逐行核对,只录入 Sonnet 4.6 主体列。图表中其他模型仅作为比较,不作为本次主体结果。
  • Claude Sonnet 4.6 System Card · 附件 · PDF with changelog dated 2026-03-06; 135 pages; collected 2026-11-10
    公布 未确认 · 采集 · 已核查 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。p.40 Pass@k 曲线和 p.51 生命科学图没有精确标签;CharXiv 图文冲突两值保留且不排名。

配置

  • Adaptive thinking, max effort · Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise. max
    未知项 Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
  • Adaptive thinking, high effort · Adaptive thinking enabled; high effort. high
    未知项 Sampling settings are not specified in every evaluation.
  • Vending-Bench 2 max effort · Max effort as named for this benchmark. max
    未知项 The report does not state whether adaptive thinking was enabled.
  • Vending-Bench 2 high effort · High effort as named for this benchmark. high
    未知项 The report does not state whether adaptive thinking was enabled.
  • No thinking, max effort · Thinking disabled; max effort. max
    未知项 Sampling settings are not specified in every evaluation.
  • Max thinking · Max thinking setting, as named by Vals AI for Finance Agent. max
    未知项 Thinking token budget and sampling settings are not stated.
  • High thinking · High thinking setting, as named by Vals AI for Finance Agent. high
    未知项 Thinking token budget and sampling settings are not stated.
  • 64K extended thinking · Extended thinking with a 64K token budget. 思考预算 64K
    未知项 Effort setting is not stated for the long-context table results.
  • Adaptive max, without tools · Adaptive thinking enabled; max effort; no tools. max
    未知项 Default sampling settings apply unless otherwise stated.
  • Adaptive max, image-cropping tool · Adaptive thinking enabled; max effort; simple image-cropping tool available. max
    未知项 Default sampling settings apply unless otherwise stated.
  • High effort, 120K thinking budget · High effort with 120K thinking tokens; ARC Prize Foundation private dataset evaluation. high · 思考预算 120K
  • Evaluation settings not fully stated · Only settings explicitly listed in each result's conditions are known.
    未知项 Effort, sampling and tool settings are unspecified unless stated in result.conditions.
  • ARC · Max effort · ARC 图表与同表脚注明确 Max effort。 max
    未知项 未说明 thinking mode 与 token budget。

直接前代与后续发布

当前收录未确认直接前代。 官方关系依据

Improvements in consistency, instruction following, and more have made developers with early access prefer Sonnet 4.6 to its predecessor by a wide margin.

测试记录 · 31