OpenAI Developers(@OpenAIDevs)

https://t.co/tLBhO7FkJt

5.2内容质量
https://t.co/tLBhO7FkJt

TL;DR · AI 摘要

OpenAI宣布GPT-5.5发布，称其为当前最强前沿模型，提升在代理式编程、计算机操作、科学计算与网络安全等任务上的表现，但未披露技术细节、API上线时间或基准测试方法。

核心要点

  • GPT-5.5是OpenAI新发布的前沿模型，主打agentic coding与计算机使用能力
  • 在Terminal-Bench、OSWorld、CyberGym等评测中取得SOTA分数，但未说明评测可复现性
  • Codex已集成GPT-5.5并上线ChatGPT，API尚未开放，属典型产品公告而非技术文档

结构提纲

按章节快速跳转。

  1. §发布通告

    OpenAI官方账号宣布GPT-5.5正式发布，定位为最新前沿模型。

  2. ·能力宣称

    强调在agentic coding、计算机操作、科学分析和网络安全等场景的性能提升。

  3. ·基准成绩

    列出多项内部评测分数，但未提供评测设置、数据集或对比基线细节。

  4. ·Codex集成

    确认GPT-5.5已用于增强Codex，在浏览器交互、文件处理等场景落地。

  5. ·开发者现状

    API尚未开放，仅限ChatGPT/Codex体验，无技术接入路径说明。

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • GPT-5.5发布通告
    • 能力定位
      • agentic coding
      • 计算机操作
      • 科学计算
      • 网络安全
    • 落地形态
      • ChatGPT集成
      • Codex增强
      • API待发布
    • 评估信号
      • 多项闭源评测SOTA
      • Preparedness风险定级
      • 开发者实测案例

金句 / Highlights

值得收藏与分享的关键句。

  • GPT-5.5是我们的智能前沿模型，面向代理式编程、计算机使用、知识工作与科学研究。

    — 第2段

    ⬇︎ 下载 PNG𝕏 分享到 X
  • 82.7% on Terminal-Bench 2.0；78.7% on OSWorld-Verified；81.8% on CyberGym——但未说明评测是否开源或可复现。

    — 第4段

    ⬇︎ 下载 PNG𝕏 分享到 X
  • Codex已用GPT-5.5优化自身基础设施：从想法到可评测实现、实验编排、推理级优化均提速。

    — 第8段

    ⬇︎ 下载 PNG𝕏 分享到 X
  • ‘我们将其网络安全能力列为High（高风险等级）’——首次将模型能力直接映射至内部Preparedness风险分级。

    — 第10段

    ⬇︎ 下载 PNG𝕏 分享到 X
#OpenAI#GPT-5.5#Codex#AI模型#开发者工具
打开原文

April kept the changelog busy. Here’s what changed for developers building with OpenAI:

It’s 5/5, so yes, GPT-5.5 gets the first slot:

Image 1: Square profile picture
Image 1: Square profile picture

OpenAI Developers

Image 2
Image 2

@OpenAIDevs

GPT-5.5 is here. It’s our smartest frontier model yet, introducing a new class of intelligence for agentic coding, computer use, knowledge work, and scientific research. Rolling out in ChatGPT and Codex today. API is coming soon.

Image 3: Square profile picture
Image 3: Square profile picture

OpenAI Developers

Image 4
Image 4

@OpenAIDevs

Replying to

GPT-5.5 reaches state-of-the-art results across key evals for agentic coding, computer use, tool use, advanced math, and cybersecurity tasks. 82.7% on Terminal-Bench 2.0 78.7% on OSWorld-Verified 55.6% on Toolathlon 35.4% on FrontierMath Tier 4 81.8% on CyberGym

Image 5: Image
Image 5: Image
Image 6: Square profile picture
Image 6: Square profile picture

OpenAI Developers

Image 7
Image 7

@OpenAIDevs

Replying to

GPT-5.5 is stronger on scientific and technical research workflows. It reaches 25.0% on GeneBench, up from 19.0% for GPT-5.4, on multi-stage scientific data analysis in genetics and quantitative biology. On FrontierMath Tier 4, it reaches 35.4%, up from 27.1% for GPT-5.4.

Image 8: Square profile picture
Image 8: Square profile picture

OpenAI Developers

Image 9
Image 9

@OpenAIDevs

Replying to

GPT-5.5 helped improve the infrastructure that serves it. To hit GPT-5.4 latency, the team used Codex and GPT-5.5 to move faster from idea to benchmarkable implementation, wire up experiments, and find inference-level optimizations. Codex analyzed weeks of production traffic

Image 10: Square profile picture
Image 10: Square profile picture

OpenAI Developers

Image 11
Image 11

@OpenAIDevs

Replying to

GPT-5.5 is a significant step up on cybersecurity task performance. It reaches 81.8% on CyberGym, up from 79.0% for GPT-5.4, and 88.1% on an expanded set of hard Capture-the-Flags challenge tasks. We’re treating its cybersecurity capabilities as High under our Preparedness

Our community is putting GPT-5.5 to work:

Diego | AI Image 12: 🚀 - e/acc

@diegocabezas01

GPT-5.5 xhigh in Codex: Look up online, everything you can find about GPT-5.5. Then give me some slides a presentation.

Image 13
Image 13

Derya Unutmaz, MD

@DeryaTR_

Codex GPT-5.5 now has fantastic frontend and design taste! This had been my main issue pre 5.5, but it is dramatically better now. I built this nearly 100,000-gene-mutation cancer site with GPT-5.5. It has beautiful UI/UX & tons of useful information! er-mutation-research-atlas.vercel.app

Image 14: Image
Image 14: Image

Peter Yang

@petergyang

I've been doing the F-Zero test each time a new model comes out and GPT 5.5 and Codex is the only combo that built a working game so far. Even made a bunch of other bots to race against. What a insane time to be building.

Image 15
Image 15

Codex got more plugins to work with your go-to tools:

Image 16: Square profile picture
Image 16: Square profile picture

OpenAI Developers

Image 17
Image 17

@OpenAIDevs

With GPT-5.5, Codex now gets more of the job done across the browser, files, docs, and your computer. We've expanded browser use so Codex can interact with web apps, and test flows, click through pages, capture screenshots, and iterate on what it sees until it completes the

Image 18
Image 18
Image 19: Square profile picture
Image 19: Square profile picture

OpenAI Developers

Image 20
Image 20

@OpenAIDevs

Replying to

Codex now generates higher-quality spreadsheets, slide decks, and documents with GPT-5.5 in

Office and

. A new file viewer in the app makes it faster to review, revise, and iterate, so files are ready to share sooner.

Image 21
Image 21
Image 22: Square profile picture
Image 22: Square profile picture

OpenAI Developers

Image 23
Image 23

@OpenAIDevs

Replying to

Computer use just got better too. With GPT-5.5, Codex is stronger at using apps on your computer, from seeing what’s on screen to clicking, typing, navigating, and moving context across tools.

Image 24
Image 24
Image 25: Square profile picture
Image 25: Square profile picture

OpenAI Developers

Image 26
Image 26

@OpenAIDevs

Auto-review is a new mode that lets Codex work longer with fewer approvals and safer execution. It helps Codex keep moving through tests, builds, and more, including during long tasks and automations, while a separate agent checks higher-risk steps in context before they run.

Image 27
Image 27

Chronicle helps Codex pick up where you left off:

Image 28: Square profile picture
Image 28: Square profile picture

OpenAI Developers

Image 29
Image 29

@OpenAIDevs

Last week, we released a preview of memories in Codex. Today, we’re expanding the experiment with Chronicle, which improves memories using recent screen context. Now, Codex can help with what you’ve been working on without you restating context.

Image 30
Image 30
Image 31: Square profile picture
Image 31: Square profile picture

OpenAI Developers

Image 32
Image 32

@OpenAIDevs

Replying to

Chronicle runs background agents to build memories from screen captures. It uses rate limits quickly. Screen captures are stored temporarily on device to generate memories—also stored on device. You can inspect and edit memories. Be aware that other apps may access these files.

Bring your setup and your team to Codex:

Image 33: Square profile picture
Image 33: Square profile picture

OpenAI Developers

Image 34
Image 34

@OpenAIDevs

Bring Codex to your team without fixed seat costs. We’re rolling out usage-based pricing for Codex in ChatGPT Business and Enterprise plans, so teams have a more flexible way to get started.

Image 35
Image 35
Image 36: Square profile picture
Image 36: Square profile picture

OpenAI

@OpenAI

Bring your workflow to Codex in just a few clicks. Import settings, plugins, agents, project configuration, and more so you can keep working with fewer interruptions. Your move.

Image 37
Image 37
Image 38: Square profile picture
Image 38: Square profile picture

OpenAI Developers

Image 39
Image 39

@OpenAIDevs

Add Codex seats with a $0 seat fee for a limited time. Through the end of June, eligible ChatGPT Business and Enterprise customers can add Codex-only seats, making it easier to give more developers access to Codex in their day-to-day workflows.

The Agents SDK added more control for long-running agents:

Image 40: Square profile picture
Image 40: Square profile picture

OpenAI Developers

Image 41
Image 41

@OpenAIDevs

Build long-running agents with more control over agent execution. New capabilities in the Agents SDK: • Run agents in controlled sandboxes • Inspect and customize the open-source harness • Control when memories are created and where they’re stored

Image 42: Image
Image 42: Image
Image 43: Square profile picture
Image 43: Square profile picture

OpenAI Developers

Image 44
Image 44

@OpenAIDevs

Replying to

Bring your own environment. The Agents SDK now supports sandbox execution with providers including

,

,

,

, and more. Keep files, credentials, and execution state in your environment while passing approved context to the model.

Image 45: Image
Image 45: Image
Image 46: Square profile picture
Image 46: Square profile picture

OpenAI Developers

Image 47
Image 47

@OpenAIDevs

Replying to

Improve agent performance with a harness that keeps long-running agents on track. It manages the agent loop across tools, context, and traces. The sandbox preserves working state across pauses, retries, and resumptions.

Image 48: Image
Image 48: Image
Image 49: Square profile picture
Image 49: Square profile picture

OpenAI Developers

Image 50
Image 50

@OpenAIDevs

Replying to

Customize the Agents SDK to fit your stack. Use it out of the box, or adapt the models, tools, instructions, and orchestration logic your agent needs. The Agents SDK capabilities are available to all API customers.

Building with TypeScript?

Image 51: Square profile picture
Image 51: Square profile picture

OpenAI Developers

Image 52
Image 52

@OpenAIDevs

The updated Agents SDK is now available in TypeScript, with support for sandbox agents and an open-source harness built in.

Quote

Image 53: Square profile picture
Image 53: Square profile picture

OpenAI Developers

Image 54
Image 54

@OpenAIDevs

Apr 15

Build long-running agents with more control over agent execution. New capabilities in the Agents SDK: • Run agents in controlled sandboxes • Inspect and customize the open-source harness • Control when memories are created and where they’re stored

Image 55: Image
Image 55: Image

We also talked to our sandbox partners

,

, and

about Agents SDK:

Image 56: Square profile picture
Image 56: Square profile picture

OpenAI Developers

Image 57
Image 57

@OpenAIDevs

With the Agents SDK and

Sandbox, agents can execute work in isolated environments while keeping credentials separate from the harness.

Image 58
Image 58
Image 59: Square profile picture
Image 59: Square profile picture

OpenAI Developers

Image 60
Image 60

@OpenAIDevs

“People aren’t just building for humans anymore. They’re building for agents.”

shares how Cloudflare Sandbox SDK works with the OpenAI Agents SDK to help agents run code in secure environments while keeping sensitive data separate from execution.

Image 61
Image 61
Image 62: Square profile picture
Image 62: Square profile picture

OpenAI Developers

Image 63
Image 63

@OpenAIDevs

Agents that run code need a controlled workspace ready when work starts.

shares why scale matters for long-running agents built with the Agents SDK.

Image 64
Image 64

WebSockets came to the Responses API:

Image 65: Square profile picture
Image 65: Square profile picture

OpenAI Developers

Image 66
Image 66

@OpenAIDevs

Image 67: ⚙️ We made agent loops faster with WebSockets in the Responses API As Codex got faster, the bottleneck moved from inference to inefficient API calls WebSockets keep response state warm across tool calls, helping workflows run up to 40% faster end to end

Symphony turns issue queues into agent workflows:

Image 68: Square profile picture
Image 68: Square profile picture

OpenAI Developers

Image 69
Image 69

@OpenAIDevs

Image 70: 📣 What if every open issue had a Codex agent? That’s the idea behind Symphony, an open-source agent orchestrator for Codex that turns task trackers into always-on systems for agentic work, letting humans focus on review and direction.

Image 71
Image 71

Alex Kotliarskyi Image 72: 🇺🇦

@alex_frantic

Engineers at OpenAI experience the same problem as everyone else — we can supervise about 3–5 coding agents. After that productivity drops. Codex is smart, but our attention is limited. So we built (and open sourced!) Symphony to remove that ceiling. Here’s how it works:

Quote

Image 73: Square profile picture
Image 73: Square profile picture

OpenAI Developers

Image 74
Image 74

@OpenAIDevs

Apr 27

Image 75: 📣 What if every open issue had a Codex agent? That’s the idea behind Symphony, an open-source agent orchestrator for Codex that turns task trackers into always-on systems for agentic work, letting humans focus on review and direction.

Image 76
Image 76

Create and edit images in Codex and the API:

Image 77: Square profile picture
Image 77: Square profile picture

OpenAI Developers

Image 78
Image 78

@OpenAIDevs

gpt-image-2 is here, available today in the API and Codex. The most capable image generation model yet, built for production-grade workflows with stronger text rendering, layout, editing, resolution, and multilingual rendering.

Image 79
Image 79
Image 80: Square profile picture
Image 80: Square profile picture

OpenAI Developers

Image 81
Image 81

@OpenAIDevs

Replying to

Create assets for the surface you need. gpt-image-2 supports thousands more export ratios and higher-resolution outputs up to 2K, making it easier to create images for apps, ads, social placements, and product workflows.

Image 82: Image
Image 82: Image
Image 83: Square profile picture
Image 83: Square profile picture

OpenAI Developers

Image 84
Image 84

@OpenAIDevs

Replying to

Text-heavy visuals get more practical. gpt-image-2 improves multilingual text rendering and structured image generation for diagrams, infographics, charts, comics, and multi-panel scenes.

Image 85: Image
Image 85: Image
Image 86: Image
Image 86: Image

People are turning gpt-image-2 into visual workflows:

Gregory Wieber

@dreamwieber

Alright, Codex with GPT 5.5 is completely cracked. This is nuts Image 87: 🤯 Basically one-shotted my request to create an app that takes a prompt, creates an equirectangular panorama with GPT image 2 – and then use Apple's ML Sharp to stitch a gaussian splat world together.

Image 88
Image 88

underpaid mom

@underpaid_mom

Time travel is now possible with gpt-image 2 and codex 5.5. WenWare is a time travel geoguessr inspired game where one can explore immersive 360° historical scenes. Created with #threejs for #vibejam 2026

Image 89
Image 89

Kris Kashtanova

@icreatelife

You can improve your equirectangular panorama video by automating your rotation in Codex + GPT Image 2 prompt: make 3d view of this panorama, make this a smooth automatic pan of the 360 view Welcome to Mars! Original tutorial and steps below

Image 90
Image 90

Quote

Kris Kashtanova

@icreatelife

Apr 22

Image 91: Embedded video
Image 91: Embedded video

0:30

Tutorial: how to make equirectangular panorama with GPT Image 2 + Codex Step 1: generate an image with GPT Image 2 prompt: make equirectangular panorama of [PLACE] Step 2: feed your panorama to Codex as a reference and prompt: make mouse-controlled 3D view I'd love to see

Build interactive voice apps:

Image 92: Square profile picture
Image 92: Square profile picture

OpenAI Developers

Image 93
Image 93

@OpenAIDevs

You can build interactive applications with gpt-realtime-1.5, so users can control app state more naturally with voice. Hi Chappy Image 94: 👋

Image 95
Image 95
Image 96: Square profile picture
Image 96: Square profile picture

OpenAI Developers

Image 97
Image 97

@OpenAIDevs

When your voice agent debugs your slides live

is using gpt-realtime-1.5

Image 98
Image 98

A lot has shipped, and the stack keeps moving.

Follow

on X to stay up to date.