Claude Haiku 5.5 · Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall average · 成绩与配置
Claude Haiku 5.5 的 Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall average 记录。本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录),附 1 条来源定位。
Claude Haiku 5.5 · Multi-task Indic Language Understanding Benchmark 成绩与配置
指标与成绩
配置与评测条件
出处
Average accuracy87.6%原文 87.6% · 单位 %
公布 未确认
System Card 标准 max effort
设置
Unless otherwise noted: adaptive thinking at max effort, default temperature/top_p, five-trial average; context window is evaluation-dependent and no more than 1M tokens.
未知项
每个 benchmark 的具体 context window 未逐项列明。
工具
禁止工具
推理预算
max
框架
未记录
协议
MILU; average accuracy over 11 languages; no tools or customized system prompt; mean of five trials.
条件
Max effort; safety classifiers enabled; blocked or unanswered items (<0.2%) were excluded rather than marked incorrect.
Claude Haiku 5.5 · Multi-task Indic Language Understanding Benchmark · 10 Indic languages plus English · overall average · Scores and configurations
Claude Haiku 5.5 record for 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). Includes 1 source location.
Claude Haiku 5.5 · Multi-task Indic Language Understanding Benchmark scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
Average accuracy87.6%Reported 87.6% · Unit %
Published Unconfirmed
System Card 标准 max effort
Settings
Unless otherwise noted: adaptive thinking at max effort, default temperature/top_p, five-trial average; context window is evaluation-dependent and no more than 1M tokens.
Unknowns
每个 benchmark 的具体 context window 未逐项列明。
Tools
Tools forbidden
Reasoning budget
max
Framework
not recorded
Protocol
MILU; average accuracy over 11 languages; no tools or customized system prompt; mean of five trials.
Conditions
Max effort; safety classifiers enabled; blocked or unanswered items (<0.2%) were excluded rather than marked incorrect.