ChatGPT & Codex in Detail

[ikaijua]

ChatGPT & Codex in Detail

ChatGPT is OpenAI's AI assistant; Codex is its AI coding agent.

One account. Subscription, quotas, and active sessions are shared under a single ChatGPT account — you can pick a task back up on phone, desktop, or web.
Division of labour. ChatGPT covers Q&A, writing, multimodal work, and everyday tasks; Codex goes into your local repo to read, edit, and run commands.

Details are split across the floors below — ChatGPT first, then Codex.

Table of Contents

ChatGPT

1. What is ChatGPT?
2. Quick Start
3. What It Can Do
4. Typical Use Cases
5. Advanced: Which Model to Choose?
6. Pricing
7. How to Check Token / Usage Limits
8. FAQ
9. Links

Codex

10. What is Codex?
11. Core ideas
12. Where you can use it
13. Desktop app (Codex App)
14. Codex Web & Codex Cloud
15. IDE extension & mobile app
16. Core capabilities
17. Quick start
18. Approval modes and sandbox
19. Which model for which task
20. Best practices
21. Related resources
22. License

Click a link to jump to that floor; if in-page clicks do nothing, open the link in a new tab.

tag: AI Assistant / Chatbot / OpenAI

[ikaijua]

What is ChatGPT?

ChatGPT is OpenAI's AI assistant, built on the GPT family of large language models. Think of it as a conversational, multi-modal digital assistant that can chat, write code, draft content, analyze data, generate images, and even remember your preferences across conversations.

The current flagship model is GPT-6 Astra. In September 2026 OpenAI also released GPT-6 Sol (near-Astra quality at ~1/5 the API price) and GPT-6 Luna (a fast decision/routing model).

[ikaijua]

Quick Start

1. Open the web app: https://chatgpt.com
2. Sign up / log in: Email, Google, Microsoft, or Apple accounts are supported.
3. Pick a plan:
   • Free: basic chat, web browsing, limited access to Sol/Luna.
   • Plus ($20/mo): higher limits, priority access to new features, GPT-6 Astra access.
   • Pro ($200/mo): higher quotas, Deep Research, Pro 500 speed tier.
   • Team / Enterprise: shared workspaces, admin controls, stronger privacy.
4. Install an app: iOS, Android, macOS, and Windows clients are available; history syncs across devices.
5. Start typing: describe what you want, e.g. "Write a polite email declining a meeting invitation." ChatGPT will give you a usable draft immediately.

Biggest gotcha: the free tier has usage caps and may queue during peak hours. For work or coding, Plus or above is much more reliable.

[ikaijua]

What It Can Do

• Conversational AI: multi-turn dialogue with strong context tracking and persistent memory.
• Coding assistant: write, explain, debug, refactor, and test code in many languages.
• Content creation: emails, reports, marketing copy, outlines, creative writing, translations.
• Multimodal understanding: analyze images, PDFs, audio, and screenshots.
• Image generation: DALL·E and GPT Image for text-to-image, inpainting, and sketch-to-image.
• Web browsing & tools: search the web, run Python, perform data analysis.
• Custom GPTs: build specialized assistants for repeated workflows.
• Memory & user profile: remembers your preferences across sessions for more personalized replies.
• Voice mode: hands-free spoken conversations with interruption support.
• Deep Research: multi-source research and synthesis for complex topics.
• Canvas / Projects / Scheduled tasks: long-document collaboration, project files, and recurring automations.
• Dots (DevDay 2026): always-on cloud agents that can handle email, calendars, and spreadsheets within permission boundaries.

[ikaijua]

Typical Use Cases

Scenario	Example prompt	Expected result
Daily Q&A	"Explain Transformers like I'm five."	A simple, analogy-rich explanation
Coding	"Write a Python script to rename all .jpg files by EXIF date."	Runnable code + explanation
Editing	"Make this paragraph more formal, under 200 words."	A polished rewrite
Reading PDFs	Upload a paper, ask "What is the core innovation in the methods section?"	Key points summarized
Image generation	"A cat in a suit working on a laptop at a café, Ghibli style"	Generated image matching the prompt
Data analysis	Upload a CSV, ask "Top 3 categories by sales and trend"	Analysis + visualization suggestions
Learning	"Give me 5 medium-difficulty Python interview questions with answers."	Questions + explanations

[ikaijua]

Advanced: Which Model to Choose?

Model	Positioning	Best for	Pricing
GPT-6 Astra	Flagship	Complex reasoning, long docs, high-quality creation	Plus/Pro, expensive API
GPT-6 Sol	High-value	Everyday chat, coding, mid-complexity tasks	~1/5 of Astra API cost
GPT-6 Luna	Fast decisions	Classification, routing, simple multi-choice	Cheapest, high-throughput
GPT-4o / 4o-mini	Legacy	Simple tasks, low-cost scenarios	Free / low-cost

Most users can leave it on Auto and let the system route to the right model.

[ikaijua]

Pricing

Plan	Monthly price	Key benefits
Free	$0	Basic chat, limited Sol/Luna, browsing, limited image generation
Plus	$20	Astra access, higher message limits, priority new features
Pro	$200	Higher quotas, Pro-only speed, priority support, more Deep Research
Team	$25/user+	Shared workspace, admin controls, stronger privacy
Enterprise	Custom	SSO, audit logs, custom terms, dedicated support

API is billed separately by token usage; Sol/Luna are significantly cheaper than Astra.

[ikaijua]

How to Check Token / Usage Limits

Web and App Users

• Check usage: Click your profile (bottom-left) → Settings → Usage to see your current cycle's percentage used, remaining quota, and next reset time.
• Plan info: Settings → Subscription shows your current plan (Free / Plus / Pro / Team) and upgrade options, but not detailed usage.
• Activity: Settings → Data controls → Export data lets you view/export conversation history.
• Limit notes: Each tier has separate caps for GPT-6 Astra, Sol, Luna, voice, image generation, and Deep Research. When you hit a cap, you can usually fall back to a cheaper model or wait for the next billing cycle to reset.

API Developers

• Usage dashboard: https://platform.openai.com/usage
• Filter by date, model, API key, and project to see token consumption and cost.
• You can set monthly billing limits and low-balance alerts to avoid runaway spend.
• Billable items include input tokens, output tokens, and cached input tokens; model prices vary widely, with Sol/Luna much cheaper than Astra.

Quick Estimation Tip

For casual chat, the web app's "message percentage" limit is easier to watch than tokens. If you are integrating via the API, focus on the usage page's total tokens and cost breakdown by model to optimize spend.

[ikaijua]

FAQ

Q: How does ChatGPT compare to Claude or Gemini?
A: ChatGPT leads in balanced general capabilities and mature persistent memory; Claude is strong at long-context coding; Gemini integrates most deeply with Google services.

Q: Is my data safe?
A: Free/Plus conversations may be used for model improvement (can be disabled). Team/Enterprise plans offer no-training and audit features. Avoid pasting sensitive data unredacted.

Q: Is ChatGPT available in China?
A: chatgpt.com is not directly accessible from mainland China without network workarounds; OpenAI does not currently offer service in mainland China.

Q: Is the free tier enough?
A: For light Q&A and simple writing, yes. For coding, heavy usage, or image generation, Plus is recommended.

Q: How do I make it remember my preferences?
A: Turn on Memory in settings; ChatGPT will learn and store facts across chats. You can also add explicit custom instructions.

Q: What is the difference between the web app and the API?
A: The web app is the end-user chat product. The API is for developers who want to integrate GPT capabilities into their own apps.

[ikaijua]

Links

• Website: https://chatgpt.com
• OpenAI API: https://platform.openai.com
• OpenAI docs: https://platform.openai.com/docs
• DevDay 2026 recap: https://openai.com/index/devday-2026/

[ikaijua]

What is Codex?

Codex is OpenAI's AI coding agent: an open-source (Apache-2.0) terminal CLI written in Rust, also available as IDE extensions (VS Code / Cursor / Windsurf / JetBrains), a desktop app, and Codex Web (cloud tasks). Active sessions sync across phone, desktop, and web via your ChatGPT account.

[ikaijua]

Core ideas

Terminal-native

Codex CLI is built around the terminal: explore, edit, run, and review inside the shell you already work in.

Local-first, cloud optional

It runs on your machine by default, so code and credentials stay local. You can also hand a task to Codex Cloud and let it work in a remote sandbox while you do something else.

Automation you can dial in

Approval modes decide when Codex pauses to ask; sandbox modes decide what it is allowed to touch — files, commands, network.

Extensible

Skills (reusable instructions), Plugins (team tool connections), MCP servers, subagents, and Hooks.

[ikaijua]

Where you can use it

Surface	Notes
Terminal CLI	Open-source Rust client — the most complete experience
IDE extensions	VS Code / Cursor / Windsurf / JetBrains
Desktop app	codex app — visual management, multiple sessions
Codex Web	chatgpt.com/codex — cloud tasks and collaboration
Mobile	The ChatGPT app can connect to and continue active sessions

[ikaijua]

Core capabilities

Understands and edits code

• Maps project structure, dependencies, and call relationships
• Fixes bugs, refactors, and adds features
• Runs multi-step tasks and iterates on the result

Command line and CI

• Interactive TUI plus one-shot prompt mode
• codex exec for scripts and pipelines
• Pipes and shell-script integration

Code review

• codex review reviews uncommitted changes, a commit, or a branch
• Leaves the working tree untouched — it reports risks and suggestions only

Extension system

• Skills / Plugins: reusable instructions and team tool connections
• MCP servers: connect external tools and data sources
• Subagents: split a large investigation across specialised agents
• Hooks: custom lifecycle behaviour

Cross-device and remote

• ChatGPT relay: active sessions follow you between phone, desktop, and web
• Remote SSH: work directly against a remote dev environment
• Codex Cloud: submit a task to the cloud and keep going with your laptop closed

[ikaijua]

Quick start

Install

Official install script:

# macOS / Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

# Windows PowerShell
powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

Homebrew:

brew install --cask codex

npm (alternative):

npm install -g @openai/codex

Verify:

codex --version

Sign in

codex

Choose Sign in with ChatGPT (recommended), or use an OpenAI API key.

Commands you'll actually use

codex                                      # start an interactive session
codex "explain the structure of this repo" # one-shot prompt
codex exec "run the test suite"            # non-interactive / CI mode
codex review                               # review uncommitted changes
codex resume                               # resume the most recent session
codex cloud                                # manage cloud tasks
codex mcp list                             # list configured MCP servers
codex --image error.png "how do I fix this" # pass an image as context

[ikaijua]

Approval modes and sandbox

Approval mode controls when Codex stops to ask you; sandbox mode controls its file and network reach.

Approval modes

Mode	Behaviour
suggest (default)	Asks before every change
auto-edit	Edits files automatically, asks before running commands
full-auto	Runs edits and commands without asking (use with care)

codex --approval-mode auto-edit
codex --approval-mode full-auto

Sandbox modes

Mode	Behaviour
read-only	Read only, no file changes (planning mode)
workspace-write	Read and write inside the current working directory
danger-full-access	No sandbox restrictions

codex -s read-only
codex --sandbox workspace-write

Switch on the fly inside the session with /permissions. In a repo you don't know yet, start with suggest plus read-only / workspace-write.

[ikaijua]

Which model for which task

Codex CLI takes a model per run via -m / --model, and a default in ~/.codex/config.toml. Models move fast — the mapping below is about complexity, how deterministic the task is, and cost sensitivity.

Currently recommended models

Model	Positioning	Best for	Cost / speed
gpt-6-astra	Highest capability	The hardest end-to-end work, complex architecture, critical judgement, a first deep pass over an unfamiliar repo	Highest cost, slowest
gpt-6.1-sol	Best value	Repetitive long-horizon work, complex coding and agentic flows — the everyday default	Close to Astra, cheaper
gpt-6-luna	Fast and cheap	Well-defined, repeatable, high-volume work: single-file fixes, filling in tests, extraction / conversion / classification	Fastest, cheapest

Available models roll out per plan and client. Run codex --help or check the Codex model docs for the current list.

How to choose

1. Luna for clear, repeatable tasks — single-file edits, bug fixes, filling in tests, formatting, structured extraction.
2. 6.1 Sol for long, complex work — cross-file refactors, feature implementation, code review, multi-step reasoning.
3. Astra for the hardest, most critical work — system architecture, hard algorithms, security-critical code, first deep analysis of an unfamiliar repo.
4. Luna for cost-sensitive batches — CI scripts, bulk refactors, repetitive edits.
5. Scout before you commit — on an unfamiliar repo, read-only + Luna for a first pass, then decide whether to escalate to Sol or Astra for the actual changes.

Examples

# Clear, repeatable everyday task
codex -m gpt-6-luna "Add unit tests for this function"

# Complex refactor that must update every call site
codex -m gpt-6.1-sol "Refactor the string helpers under utils/ into their own module and update all call sites"

# First deep analysis of an unfamiliar repo
codex -m gpt-6-astra -s read-only "Map module dependencies across src/ and flag circular dependencies and design risks"

# Per-session override via environment variable
OPENAI_CODEX_MODEL=gpt-6.1-sol codex "Implement the user login endpoint"

Setting a default model

model = "gpt-6.1-sol"

With no model set, Codex CLI picks a recommended model for your account and client. The config format can change between CLI versions — check codex --help or the official docs after setting one.

[ikaijua]

Best practices

1. Start in suggest mode — raise automation only once you know the repo.
2. State the scope — name the files, the expected behaviour, and the tests to run.
3. Review the diff and run the tests — AI changes still need a human check.
4. Capture repeat work as Skills — stop re-typing the same prompt.
5. Sandbox sensitive code tightly — avoid accidental file or network access.

[ikaijua]

Related resources

• GitHub repository
• Codex CLI docs
• Codex model docs
• OpenAI announcement: work with Codex from anywhere
• Chinese version, maintained side by side: ChatGPT & Codex in Detail (Chinese)

[ikaijua]

License

Codex CLI is released under the Apache-2.0 license. Using OpenAI's model service is subject to the corresponding terms of service.

[ikaijua]

Desktop app (Codex App)

The Codex desktop app (macOS Feb 2026, Windows Mar 2026) is a command center for agents.

Built to manage many coding agents at once — not a single-threaded chat.

Parallel agents. Agents run in per-project threads; run many tasks side by side and switch without losing context.
Worktree isolation. Built-in git worktrees: each agent works on an isolated copy, so multiple agents on one repo don't conflict; check out changes locally or let the agent continue without touching your local git state.
Diff review. Inspect agent changes in the thread, comment on diffs, or open files in your editor for manual edits.
Skill management. A dedicated interface to create and manage skills (bundled instructions, resources, scripts) — ask Codex to use specific skills or let it pick automatically per task.
Automations. Scheduled background work (daily issue triage, CI-failure digests, release briefs) lands in a review queue for you to pick up.
Personality. /personality toggles between terse/pragmatic and conversational/empathetic styles (app, CLI, IDE) — capabilities unchanged.
Session inheritance. Picks up session history and config from the Codex CLI and IDE extension, so existing projects carry over seamlessly.

Advanced capabilities (rolling out since Apr 2026):

• In-app browser: comment directly on pages for precise visual feedback — great for frontend and game iteration
• Background computer use (macOS): Codex can see, click, and type in other apps on your Mac; multiple agents run in parallel without interfering with your own work
• 90+ plugins: Atlassian Rovo (JIRA), CircleCI, CodeRabbit, GitLab Issues, Microsoft Suite, Neon, Remotion, Render, Superpowers, and more
• Goal mode: set an outcome and success criteria, and let Codex keep working toward it (app / IDE / CLI)
• Appshots (macOS): hotkey-attach a screenshot plus available text of an app window to a thread — no long setup prompt
• Memory preview: Codex remembers your preferences, corrections, and hard-won context so future tasks start faster

Security. Native system-level sandboxing: agents edit only their working folder/branch and use cached web search by default; elevated commands (e.g. network access) require permission; project/team rules can auto-allow specific commands.

[ikaijua]

Codex Web & Codex Cloud

Codex Web (chatgpt.com/codex) is the in-browser entry to cloud tasks — no local CLI install needed.

Codex Cloud runs coding tasks on OpenAI-managed machines in reusable sandbox environments:

• Reusable environments: create an environment once per repo (connect the repo, configure credentials and network access), then reuse it for later tasks
• Keeps going when your laptop closes: cloud tasks continue after sleep or lid-down
• Continue on any device: environments are created/published from the desktop app or web, then started and resumed from desktop, web, or the mobile app
• Mobile remote: the Remote tab in the mobile app gives access to desktop Codex sessions

Best for: skipping local setup, long-running tasks, and picking work up on another device.

[ikaijua]

IDE extension & mobile app

IDE extensions (VS Code / Cursor / Windsurf / JetBrains)

• Chat and edit code right inside the editor, review inline diffs
• Sessions sync with the CLI and desktop app
• Goal mode works across app / IDE / CLI

Mobile (ChatGPT app, Codex preview since Sep 2026)

• iOS & Android, all plans including Free and Go
• Connect by scanning the QR code from the Codex app on a macOS host
• Start or continue threads from your phone: answer questions, redirect, approve actions, review findings, even switch between connected hosts
• Loads live host context: project context, approvals, plugins, screenshots, terminal output, diffs, and test results
• Prerequisite: the host stays awake, online, and running Codex; macOS hosts only for now