DataLearner 标志

Claude Sonnet 4.6vsClaude Sonnet 4.5

在 13 个共同 benchmark 中,Claude Sonnet 4.6 整体领先:Claude Sonnet 4.6 领先 11 项,Claude Sonnet 4.5 领先 2 项,持平 0 项,平均分差 +21.90。

Anthropic
Claude Sonnet 4.6

Anthropic · 2026-02-17 · 聊天大模型

Anthropic
Claude Sonnet 4.5

Anthropic · 2025-09-30 · 聊天大模型

Claude Sonnet 4.611 (85%)(15%)2 Claude Sonnet 4.5

评测分数

按能力类目分组,每组内按分差大小排列;共 13 项。

Agent Level Benchmark

Claude Sonnet 4.6 领先 2/2
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
Terminal Bench Hard46.2029 / 244Normal (With Tools)28.80116 / 244Normal (With Tools)+17.40
τ²-Bench - Telecom79.50115 / 264Normal (With Tools)70.50138 / 264Normal (With Tools)+9

AI Agent - Tool Usage

Claude Sonnet 4.6 领先 2/2
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
Terminal Bench 2.059.1022 / 48Thinking (With Tools)42.8043 / 48Thinking (With Tools)+16.30
OSWorld-Verified72.5017 / 26Thinking (With Tools)61.4022 / 26Thinking (With Tools)+11.10

AI Agent - Information Search

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
BrowseComp74.7029 / 57Thinking (With Tools)24.1055 / 57Thinking (With Tools)+50.60

Claw-style Agent Evaluation

Claude Sonnet 4.5 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
Pinch Bench886 / 38Thinking (With Tools)88.205 / 38Thinking (With Tools)-0.20

General Knowledge

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
HLE13.30355 / 563Normal (No Tools) · Text only7.20436 / 563Normal (No Tools) · Text only+6.10

Instruction Following

Claude Sonnet 4.5 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
IF Bench42.40201 / 282Normal (No Tools)42.70199 / 282Normal (No Tools)-0.30

Long Context

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
AA-LCR69.30107 / 170Normal (No Tools)54133 / 170Normal (No Tools)+15.30

Multimodal Understanding

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
MMMU-Pro70.60123 / 227Normal (No Tools)65.20151 / 227Normal (No Tools)+5.40

Productivity Knowledge

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
GDPval-AA575 / 15Thinking (No Tools)3910 / 15Thinking (No Tools)+18

Writing and Creative Capabilities

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
Creative Writing1,80418 / 106Normal (No Tools)1,67430 / 106Normal (No Tools)+129.80

科学与综合推理

Claude Sonnet 4.6 领先 1/1
评测项Claude Sonnet 4.6Claude Sonnet 4.5分差
GPQA Diamond79.90237 / 462Normal (No Tools)73.70292 / 462Normal (No Tools)+6.20

规格对比

字段Claude Sonnet 4.6Claude Sonnet 4.5
发布机构AnthropicAnthropic
发布时间2026-02-172025-09-30
模型类型聊天大模型聊天大模型
架构稠密模型稠密模型
参数规模暂无数据暂无数据
上下文长度1M1000K
最大输出8K64K

API 调用价格

价格优先使用 DataLearner 配置的 API 记录;缺失项不做推测。

价格项Claude Sonnet 4.6Claude Sonnet 4.5
文本输入$3 / 1M tokens$3 / 1M tokens
文本输出$15 / 1M tokens$15 / 1M tokens
缓存读取$0.3 / 1M tokens$0.3 / 1M tokens
缓存写入$3.75 / 1M tokens$3.75 / 1M tokens

小结

  • Claude Sonnet 4.6在以下类目领先:Agent Level Benchmark (2/2)、AI Agent - Tool Usage (2/2)、AI Agent - Information Search (1/1)、General Knowledge (1/1)、Long Context (1/1)、Multimodal Understanding (1/1)、Productivity Knowledge (1/1)、Writing and Creative Capabilities (1/1)、科学与综合推理 (1/1)
  • Claude Sonnet 4.5在以下类目领先:Claw-style Agent Evaluation (1/1)、Instruction Following (1/1)

13 个共同 benchmark 上,Claude Sonnet 4.6 平均高出 21.90 分。

单项差距最大的 benchmark:Creative Writing — Claude Sonnet 4.6 1,804,Claude Sonnet 4.5 1,674(分差 +129.80)。

本页正文由结构化模型、价格与 benchmark 数据生成,不使用实时 LLM 撰写。