DataLearner 标志

Claude Opus 4.6vsOpus 4.5

在 14 个共同 benchmark 中,Claude Opus 4.6 整体领先:Claude Opus 4.6 领先 11 项,Opus 4.5 领先 3 项,持平 0 项,平均分差 +17.34。

Anthropic
Claude Opus 4.6

Anthropic · 2026-02-05 · 推理大模型

Anthropic
Opus 4.5

Anthropic · 2025-11-25 · 推理大模型

Claude Opus 4.611 (79%)(21%)3 Opus 4.5

评测分数

按能力类目分组,每组内按分差大小排列;共 14 项。

Agent Level Benchmark

Claude Opus 4.6 领先 2/3
评测项Claude Opus 4.6Opus 4.5分差
τ²-Bench91.891 / 44Extended (with tools)81.9913 / 44Extended (with tools)+9.90
Terminal Bench Hard48.5025 / 244Normal (With Tools)40.9050 / 244Normal (With Tools)+7.60
τ²-Bench - Telecom84.8090 / 264Normal (With Tools)86.3082 / 264Normal (With Tools)-1.50

Coding and Software Engineer

Claude Opus 4.6 领先 2/3
评测项Claude Opus 4.6Opus 4.5分差
Text Arena (Coding)1,5558 / 35Normal (No Tools)1,47920 / 35Normal (No Tools)+76.35
GSO33.335 / 21Normal (With Tools)26.508 / 21Normal (With Tools)+6.83
SWE-bench Verified80.8410 / 116Extended (with tools)80.909 / 116Extended (with tools)-0.06

General Knowledge

Claude Opus 4.6 领先 2/2
评测项Claude Opus 4.6Opus 4.5分差
ARC-AGI-19240 / 147Extended (no tools)8073 / 147Extended (no tools)+12
CritPt2.80120 / 200Normal (No Tools)0.30182 / 200Normal (No Tools)+2.50

AI Agent - Tool Usage

Claude Opus 4.6 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
Terminal Bench 2.065.4011 / 48Extended (with tools)59.3020 / 48Extended (with tools)+6.10

Instruction Following

Claude Opus 4.6 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
IF Bench44.60182 / 282Normal (No Tools)43194 / 282Normal (No Tools)+1.60

Long Context

Opus 4.5 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
AA-LCR67116 / 170Normal (No Tools)70.7099 / 170Normal (No Tools)-3.70

Multimodal Understanding

Claude Opus 4.6 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
MMMU-Pro72.50113 / 226Normal (No Tools)71.20120 / 227Normal (No Tools)+1.30

Writing and Creative Capabilities

Claude Opus 4.6 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
Creative Writing1,80419 / 106Normal (No Tools)1,68329 / 106Normal (No Tools)+120.90

科学与综合推理

Claude Opus 4.6 领先 1/1
评测项Claude Opus 4.6Opus 4.5分差
GPQA Diamond84176 / 462Normal (No Tools)81220 / 462Normal (No Tools)+3

规格对比

字段Claude Opus 4.6Opus 4.5
发布机构AnthropicAnthropic
发布时间2026-02-052025-11-25
模型类型推理大模型推理大模型
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1000K200K
最大输出64K64K

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 Opus 4.6Opus 4.5
文本输入$5 / 1M tokens$5 / 1M tokens
文本输出$25 / 1M tokens$25 / 1M tokens
缓存读取$0.5 / 1M tokens$0.5 / 1M tokens
缓存写入$6.25 / 1M tokens$6.25 / 1M tokens

小结

  • Claude Opus 4.6在以下类目领先:Agent Level Benchmark (2/3)、Coding and Software Engineer (2/3)、General Knowledge (2/2)、AI Agent - Tool Usage (1/1)、Instruction Following (1/1)、Multimodal Understanding (1/1)、Writing and Creative Capabilities (1/1)、科学与综合推理 (1/1)
  • Opus 4.5在以下类目领先:Long Context (1/1)

14 个共同 benchmark 上,Claude Opus 4.6 平均高出 17.34 分。

单项差距最大的 benchmark:Creative Writing — Claude Opus 4.6 1,804,Opus 4.5 1,683(分差 +120.90)。

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。