跳转到主要内容

GPT-5.5 (Xhigh)

OpenAIGPTProprietary

描述

GPT-5.5 is OpenAI's smartest model yet, designed for real work across agentic coding, computer use, knowledge work, and early scientific research. It matches GPT-5.4 per-token latency in real-world serving while reaching a much higher level of intelligence and using significantly fewer tokens to complete the same tasks. GPT-5.5 supports a 1M-token context window in the API and a 400K-token context window in Codex, with state-of-the-art results on Terminal-Bench 2.0, OSWorld-Verified, GDPval, FrontierMath, and CyberGym.

发布日期
2026-04-23
参数规模
—
上下文长度
1.1M
支持模态
image, pdf, text

能力雷达图

70
general
70
coding
60
reasoning
77
science估算
80
agents
85
multimodal

缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。

排行榜排名

领域#排名分数来源
智能体能力模型榜58
50.0
LS
代码能力榜23
85.0
LB
数学推理8
96.0
LB
多模态榜28
61.0
LS
推理能力15
90.0
LB

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Agents

BixBench80.5%自报
Finance Agent60.0%自报
Toolathlon55.6%自报
OfficeQA Pro54.1%自报
Finance Agent v251.8%
GeneBench25.0%自报

Code

CyberGym81.8%自报
FrontierSWE73.0%
DeepSWE 1.167.0%

Communication

Tau2 Telecom98.0%自报

General

GDPval-MM84.9%自报

Legal

Legal Agent Benchmark2.1%

Long Context

MRCR v2 (8-needle)74.0%自报

Math

LiveBench80.7%
FrontierMath35.4%自报

Multimodal

OSWorld-Verified78.7%自报

Reasoning

ARC-AGI95.0%自报
GPQANYU + Cohere + Anthropic (2023)93.6%自报
ARC-AGI v285.0%自报
BrowseCompOpenAI (2025)84.4%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)82.7%自报
MCP Atlas75.3%自报
SWE-Bench ProPrinceton NLP (2024)58.6%自报
Graphwalks parents >128k58.5%自报
Humanity's Last Exam52.2%自报
Graphwalks BFS >128k45.4%自报
FrontierCode 1.143.0%

Vision

MMMU-Pro83.2%自报

AA 评测指数

(Artificial Analysis)

暂无 AA 评测数据

LLM Stats 分类评分

(LLM Stats (zeroeval))
Communication
100
Physics
90
Biology
90
Chemistry
90
Multimodal
80
Safety
80
Search
80
Tool Calling
80
Vision
80
Spatial Reasoning
70
Finance
70
General
70
Code
70
Long Context
60
Math
60
Reasoning
60
Agents
60
Science
20
Legal
0

定价

输入价格$5 / 1M tokens
输出价格$30 / 1M tokens
混合价格(3:1)$11.25 / 1M tokens
缓存读取价格$0.5 / 1M tokens

速度

Tokens/秒74.1
首Token延迟45.54s
首回答延迟45.54s

供应商价格排行

供应商价格排行

4 个供应商

最便宜: Kunavo最贵: Neon
供应商输入输出
1Kunavo最便宜
$2
$12
2OpenAI主要
$5
$30
3FrogBot
$5
$30
4Neon
$5
$30

比较该模型在不同 API 供应商之间的定价。

外部链接