DataLearner 标志

Claude Opus 5.5vsGPT-6 Sol

在 23 个共同 benchmark 中,Claude Opus 5.5 整体领先:Claude Opus 5.5 领先 23 项,GPT-6 Sol 领先 0 项,持平 0 项,平均分差 +9.31。

Anthropic
Claude Opus 5.5

Anthropic · 2026-09-22 · 推理大模型

OpenAI
GPT-6 Sol

OpenAI · 2026-09-22 · 推理大模型

Claude Opus 5.523 项(100%)(0%)0 项GPT-6 Sol

评测分数

按能力类目分组,每组内按分差大小排列;共 23 项。

金融

Claude Opus 5.5 领先 3/3
评测项Claude Opus 5.5GPT-6 Sol分差
Tax Agent Bench70.506 / 20Max (With Tools)53.0520 / 20Max (With Tools)+17.45
Finance Agent v258.598 / 42Max (With Tools)49.0531 / 42Max (With Tools)+9.54
EMB (Excel Modeling)75.942 / 42Max (With Tools)71.538 / 42Max (With Tools)+4.41

临床工作与医疗决策

Claude Opus 5.5 领先 2/2
评测项Claude Opus 5.5GPT-6 Sol分差
MedScribe91.431 / 66Max (With Tools)82.0337 / 66Max (With Tools)+9.40
MedCode49.8011 / 64Max (With Tools)47.0727 / 64Max (With Tools)+2.73

客服与业务流程

Claude Opus 5.5 领先 2/2
评测项Claude Opus 5.5GPT-6 Sol分差
Public Benefits Bench v1.170.643 / 26Max (With Tools)56.6322 / 26Max (With Tools)+14.01
SAGE45.8329 / 64Max (With Tools)44.7932 / 64Max (With Tools)+1.04

法律

Claude Opus 5.5 领先 2/2
评测项Claude Opus 5.5GPT-6 Sol分差
Legal Research Bench50.484 / 42Max (With Tools)28.8532 / 42Max (With Tools)+21.63
Harvey's Legal Agent Benchmark3.7525 / 43Max (With Tools)1.6731 / 43Max (With Tools)+2.08

程序生成与编辑

Claude Opus 5.5 领先 2/2
评测项Claude Opus 5.5GPT-6 Sol分差
Program Bench18.5011 / 15Max (With Tools)214 / 15Max (With Tools)+16.50
Vibe Code Bench v1.190.292 / 60Max (With Tools)87.826 / 60Max (With Tools)+2.47

多轮记忆与持续上下文

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
EBR-bench71.432 / 23Max (With Tools)53.334 / 23Max (With Tools)+18.10

常识推理

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
SimpleBench88.401 / 96Historical report (mode unspecified)73.1016 / 96Historical report (mode unspecified)+15.30

数学

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
ProofBench v1.1 (Lean 4)1001 / 28Max (With Tools)837 / 28Max (With Tools)+17

文档与图表理解

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
GDP.pdf26.2014 / 122Max (No Tools)24.8018 / 122Max (No Tools)+1.40

机器学习工程

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
WeirdML v30.312 / 13Extra-High (With Tools)0.204 / 13Extra-High (With Tools)+0.11

桌面工作流

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
OSWorld 2.081.801 / 13Max (With Tools)64.407 / 13Max (With Tools)+17.40

生物与基因

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
BioMysteryBench79.261 / 4Max (With Tools)74.812 / 4Max (With Tools)+4.45

科学知识与推理

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
CritPt31.712 / 204Max (No Tools)30.866 / 204Max (No Tools)+0.85

科学计算与科研编程

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
SciCode66.901 / 134Max (No Tools)57.6417 / 134Max (No Tools)+9.26

算法与竞赛编程

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
IOI (Vals v2)95.062 / 26Max (With Tools)82.617 / 26Max (With Tools)+12.45

维护、优化与工程管理

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
Code Migration66.652 / 43Max (With Tools)57.204 / 43Max (With Tools)+9.45

跨能力聚合指数

Claude Opus 5.5 领先 1/1
评测项Claude Opus 5.5GPT-6 Sol分差
Vals Index69.692 / 42Max (task-specific tools)62.579 / 42Max (task-specific tools)+7.12

规格对比

字段Claude Opus 5.5GPT-6 Sol
发布机构AnthropicOpenAI
发布时间2026-09-222026-09-22
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1M1.05M
最大输出128K128K

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 5.5GPT-6 Sol
文本输入$4 / 1M tokens$2 / 1M tokens
文本输出$20 / 1M tokens$10 / 1M tokens
缓存读取$0.2 / 1M tokens$0.2 / 1M tokens
缓存写入$5 / 1M tokens$2.5 / 1M tokens

小结

  • Claude Opus 5.5在以下类目领先:金融 (3/3)、临床工作与医疗决策 (2/2)、客服与业务流程 (2/2)、法律 (2/2)、程序生成与编辑 (2/2)、多轮记忆与持续上下文 (1/1)、常识推理 (1/1)、数学 (1/1)、文档与图表理解 (1/1)、机器学习工程 (1/1)、桌面工作流 (1/1)、生物与基因 (1/1)、科学知识与推理 (1/1)、科学计算与科研编程 (1/1)、算法与竞赛编程 (1/1)、维护、优化与工程管理 (1/1)、跨能力聚合指数 (1/1)

23 个共同 benchmark 上,Claude Opus 5.5 平均高出 9.31 分。

单项差距最大的 benchmark:Legal Research Bench — Claude Opus 5.5 50.48,GPT-6 Sol 28.85(分差 +21.63)。

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。