OpenAI’s product lineup for 2026 has expanded from a single ChatGPT into seven product tracks: language models (GPT-5.5 series), programming agents (Codex), video generation (Sora 2), browser agents (Operator), image generation (gpt-image-1), speech recognition (Whisper), and vector embeddings (Embeddings). This article summarizes the positioning, pricing, use cases, and subscription tier mappings for each product track—so you don’t have to read every OpenAI official document to know in 5 minutes how much you need to pay and what you can use.
OpenAI full product track overview (latest as of May 2026)
Product Type Purpose How to get GPT-5.5 series Language model Writing, research, programming, reasoning ChatGPT App, API Codex Programming agent Write code, modify codebase, run shell commands Codex App (desktop), CLI, IDE integrations Sora 2 Video generation Short films, ads, visual content Sora website, ChatGPT integration Operator Browser agent Operate websites on your behalf, make reservations, shop, fill out forms Included in ChatGPT Pro gpt-image-1 Image generation Product images, illustrations, community assets Included in ChatGPT, API Whisper Speech recognition Meeting transcription, subtitles, voice input API (billed per minute) Embeddings Vector embeddings RAG retrieval, semantic search, similarity API (billed per token)
GPT-5.5 series: OpenAI’s core language models
GPT-5.5 is OpenAI’s current flagship model, launched on April 24, 2026. The series is divided into three tiers:
GPT-5.5: standard version, default for ChatGPT Plus subscription, API is open
GPT-5.5 Pro: flagship version, exclusive to ChatGPT Pro ($200/month), stronger than 5.5 on complex reasoning tasks
GPT-5.5-mini: budget version, used when running large volumes of tasks via the API
Key upgrades in GPT-5.5: context increased from 200K to 1M tokens, improved accuracy for agent task tool usage, Terminal-Bench reaching 82.7% (OpenAI’s strongest benchmark for agent tasks). See the abmedia report on the GPT-5.5 launch.
There’s also GPT-5.4-Cyber (security-focused version, limited to certified defenders), and the older GPT-4o and GPT-4 Turbo can still be specified for use on the API, but they are no longer the default. For new projects, just use the 5.5 series—no need to keep old versions.
Codex: OpenAI’s programming agent
Codex is OpenAI’s programming agent built for developers, positioned similarly to Anthropic’s Claude Code. It can directly read and write local files, execute shell commands, connect to GitHub to handle PRs, and handle multi-step code refactoring tasks.
How to get Codex:
Codex desktop app: macOS/Windows, provides a graphical multi-session side-by-side layout
Codex CLI: terminal version, can integrate with shell scripts
IDE integrations: provided as extensions for VS Code and the JetBrains suite
Codex data announced in May 2026: within one week of launch, API revenue growth speed doubled; it is OpenAI’s fastest-growing product track right now (see the 5/2 report). Recent updates include “Codex Pets” (custom sprite display on the desktop—no impact on work, but adds immersion), Goblin behavior fixes, and multi-session management for enterprise users. See OpenAI’s complete Codex teaching pillar.
Sora 2: OpenAI’s video generation model
Sora 2 is OpenAI’s video generation model launched in the second half of 2025, and in 2026 it has already shipped reliably to ChatGPT Plus and Pro subscribers. It can generate high-resolution short videos lasting 5–60 seconds based on text descriptions, supports character motions, scene transitions, and camera movement/cinematography control.
How to access Sora 2:
Sora dedicated website: sora.openai.com, offers a full generation interface and a materials library
ChatGPT integration: enter a prompt directly in a ChatGPT conversation to generate short videos
API: billed based on generation seconds and resolution (details in OpenAI’s announcement)
In practice, Sora 2’s advantage is “higher accuracy in understanding text-to-video instructions”—for complex scenes (multiple characters, indoor/outdoor switching, specific camera moves), the generation accuracy is significantly improved compared to the original Sora. The downside is that character consistency for long videos (over 60 seconds) and complex physical interactions still occasionally break down. Common use cases for users in Taiwan: social media marketing shorts, product demos, and animated explainer background clips.
Operator: OpenAI’s browser agent
Operator is an agent OpenAI released for browser automation—users can give instructions to have Operator open webpages, fill out forms, make reservations, shop, retrieve data, and handle follow-up processing. Operator is built into the ChatGPT Pro ($200/month) plan; it’s not yet available for Plus users.
Typical uses:
“Book me a French restaurant in Xinyi District, Taipei next Wednesday at 7 PM” → Operator opens the browser to search, check ratings, call, or make the reservation
“Compare the prices for 5 nights next month among these 3 hotels, and put them in a table” → Operator opens Booking, Agoda, and the official websites to check one by one
“Change this S3 bucket in AWS to only allow specific IPs” → Operator executes it in the console
Difference between Operator and Codex: Codex operates on code and files, while Operator operates on “webpages and forms.” For non-engineers, Operator is a more intuitive automation option; for engineers, Codex is still the mainstay for coding.
gpt-image-1, Whisper, Embeddings: API tool lineup
In addition to consumer-side products, OpenAI also provides three API tools for developers:
gpt-image-1—image generation model for 2026, replacing DALL-E 3 as OpenAI’s core. It can generate multiple sizes such as 1024×1024, 1024×1792, and 1536×1024, with improved quality and consistency over DALL-E 3. Called directly in ChatGPT chats, and billed per number of generated images on the API. The next-generation gpt-image-2 was launched in April 2026; it requires OpenAI organization verification before you can use the API, while the ChatGPT built-in version is provided directly to subscribing users.
Whisper—speech recognition model, supports multiple languages (including Traditional Chinese, Taiwan accents), billed per minute of audio. Practical applications: automatic meeting transcription, podcast subtitles, and voice-to-text input tools. Whisper is the only product from OpenAI that publicly provides model weights (download and run self-hosted), but cloud APIs typically perform better and also come with automatic updates.
Embeddings—vector embedding model that turns text into high-dimensional vectors for RAG (retrieval-augmented generation), semantic search, and document similarity matching. OpenAI’s main endpoints are currently text-embedding-3-large and text-embedding-3-small. For developers who need to build an internal enterprise knowledge base, customer service FAQ systems, and document classification, Embeddings is core infrastructure.
OpenAI subscription tiers: which plan can use which products
Plan Monthly fee Main differences Free $0 GPT-5.5 limited, no Sora/Operator/Codex Pro Plus $20 Full GPT-5.5, Sora 2, gpt-image-1, Codex standard quota Pro $200 GPT-5.5 Pro, Operator, Codex high quota, priority compute Team $25/user Enterprise collaboration, team knowledge base, no requirement to train models guarantees Enterprise Custom pricing SSO, SOC 2, contract customization, unlimited quota
For individual users, $20 Plus covers most scenarios—only heavy users of Operator/Codex, or developers who need GPT-5.5 Pro, need to upgrade to $200 Pro. For team or company users, the “no model training” guarantee in the Team plan is a necessary condition in most compliance scenarios.
Common questions FAQ
Do I absolutely need to subscribe to OpenAI? What can the free version do?
No. The free version lets you use GPT-5.5 (with limitations), basic image generation, and voice input. But Sora, Operator, Codex advanced features, and GPT-5.5 Pro require a subscription. For most users, $20 Plus is enough.
Is ChatGPT Pro $200 worth it?
Depends on how you use it: if you write lots of code daily (heavy Codex usage), need GPT-5.5 Pro’s strongest reasoning, or need Operator to automate web work—then it’s worth it. For general writing, research, and users who just look things up, $20 Plus is enough.
OpenAI API vs ChatGPT subscription—what’s cheaper?
API is billed by token and is suited for developers who want to integrate OpenAI via code. ChatGPT subscriptions are unlimited-style per month (with usage limits), suited for users who want to work through a chat interface. Most people end up having both: subscribe to chat with ChatGPT, and use the API to connect their own tools.
Can Sora 2 be used in Taiwan?
Yes. OpenAI has no regional restrictions in Taiwan, and payment supports credit cards. Sora 2 can be used in both ChatGPT Plus and Pro.
Which is better: OpenAI Codex or Anthropic Claude Code?
Each has its strengths. Codex integrates more completely with the OpenAI ecosystem (Sora, gpt-image-1, Operator); Claude Code is often chosen by developers as the first option because it shines in CLI/desktop multi-session and strictly following instructions. If your budget allows, try both for a month and see which fits your workflow better. See the full comparison of ChatGPT vs Claude vs Gemini.
How is OpenAI’s Operator different from Anthropic’s agent?
Operator focuses on browser operations (webpages, forms, reservations), with the goal of replacing the work of a person sitting at a computer clicking with a mouse. Anthropic’s agent direction is Claude Code and Mythos—Claude Code operates on code and files, while Mythos is an automated research agent at the research preview stage. The two paths diverge; the practical choice depends on your work scenarios.
This article: OpenAI full product lineup for 2026—how to choose GPT-5.5, Codex, Sora, Operator, and subscription plans first appeared on 链新闻 ABMedia.
