OpenAI

ChatGPT

Provider

OpenAI

Price

$0-500/mo

Platforms

Web, iOS, Android

Weekly users

900,000,000

Launched

Nov 2022

Updated October 2026

ChatGPT is OpenAI’s AI assistant. Launched on 30 November 2022, it reached 1 million users in 5 days, had 800 million weekly active users by October 2025 and passed 1 billion monthly users in June 2026.

Its flagship model is GPT-6 Astra, on every paid tier (Plus, Pro, Business and Enterprise) since 5 September 2026. Regular chat moved to GPT-6 on 7 October 2026: paid tiers get GPT-6 Sol, and Free and Go users get GPT-6 Luna from 8 October. The app covers chat, research, image and video generation, real-time voice, coding and always-on agents, on plans from free to $500 a month.

Quick stats

Pricing breakdown

ChatGPT offers six tiers — Free, Go, Plus, Pro (at $100, $200 or $500), Business and Enterprise. All prices are in USD.

Free — $0

The free tier includes:

• GPT-6 Luna — the default model on the Free and Go tiers from 8 October 2026, replacing GPT-5.6 Luna (MacRumors). Unlimited text chats, subject to abuse guardrails, and a Think button for harder questions have applied since August; uploads, images, voice and other tools stay limited (OpenAI)
• GPT Store — Use others’ custom GPTs
• Web search and basic voice
• Limited file uploads and image generation
• Deep Research — A small monthly allowance

Ads: the Free and Go tiers show ads in a growing number of markets. Ads began as a US test on the Free and Go tiers on 9 February 2026 and reached Australia, New Zealand and Canada on 16 April and the UK on 4 June (OpenAI); Japan, South Korea, Mexico and Brazil followed by mid-August, which OpenAI says made nine markets in all, and 31 European countries came the following week (OpenAI; PPC Land), then India on 27 August (TechCrunch). In India more than 50 brands went live through WPP and Omnicom, with self-serve buying from 4 September. OpenAI also offers a no-ads free option with lower limits where available (OpenAI). Free users also get the most restrictive limits and slower access at peak times.

Worth it for: Casual users, students, and anyone trying ChatGPT before paying.

Go — $8/month

A cheaper paid tier introduced in 2026, sitting between Free and Plus:

• GPT-6 Luna as the default, with unlimited text chats and more messages with tools than Free
• More uploads, image generation and memory than Free
• Projects, Tasks and custom GPTs
• Can show ads, like Free (OpenAI)

Worth it for: Light-to-moderate users who want more than Free without the jump to $20.

Plus — $20/month

The entry point to the current flagship. Everything in Go, plus:

• GPT-6 Sol in regular chat from 7 October 2026 and GPT-6 Astra — the current flagship — alongside the GPT-5.6 reasoning models, and Agent Mode for multi-step tasks
• Higher limits and Deep Research
• Image generation, Advanced Voice, and Sora video
• Codex — agentic coding
• Custom GPTs, Projects and Connectors (Gmail, Calendar, Drive, GitHub)

The price has held at $20/month since 2023.

Worth it for: Most regular and professional users who want the flagship model, agents and the full feature set.

Subscribe to Plus →

Pro — $100, $200 or $500/month

Pro comes in three sizes, measured against the Plus allowance. OpenAI added the $500 plan at DevDay on 29 September 2026 and reopened the $200 plan to new subscribers with a smaller allowance (The Next Web). Everything in Plus, plus:

• Pro 100 ($100): 5x Plus usage
• Pro 200 ($200): 10x Plus usage in ChatGPT Work and Codex and 100 GPT-6 Pro messages a week. Subscribers from before DevDay keep 20x and 200 messages until 29 October and get a one-time $2,500 usage credit that expires at the end of the year
• Pro 500 ($500): 25x Plus usage and Astra Ultrafast, which generates up to 300 tokens a second in Codex, up to eight times faster than standard
• Pro reasoning powered by GPT-6 Astra — the maximum-reasoning mode — with GPT-5.6 Sol Pro still available
• One dot, an always-on agent, on every Pro plan (not available in the EEA, Switzerland or the UK; see below)
• Maximum Deep Research and Sora capacity and priority compute

Worth it for: Researchers, consultants and developers whose work depends heavily on AI for several hours a day.

Business — $25–30/user/month

Formerly called “Team” (renamed August 2025). Designed for organisations:

• GPT-6 Sol, GPT-6 Astra and the GPT-5.6 models, the same line-up as Plus
• Admin dashboard — User management, usage analytics
• Data excluded from training — Your conversations are never used to train models
• SAML SSO — Enterprise authentication
• Shared workspaces — Collaborate on projects across team members
• Priority support

Minimum 2 users. Annual billing available at reduced rates. Since August 2026 a workspace can also buy Premium seats at $125 per user a month, or $100 billed annually, with five times a standard seat’s usage; a Premium seat includes one dot (OpenAI).

Learn more about Business →

Enterprise — Custom pricing

For large organisations requiring maximum security and customisation. Estimated ~$60/user/month based on public reports:

• 2 million token context window — Handle massive documents
• Custom data retention policies
• SOC 2 Type 2 compliant
• Enterprise Key Management (EKM) — Customer-controlled encryption keys
• SCIM provisioning — Automated user management
• Data residency options — US, Europe, UK, Japan, Canada, South Korea, Singapore, Australia, India, UAE
• Dedicated support and onboarding

92% of Fortune 500 companies use ChatGPT Enterprise, according to OpenAI.

Contact sales →

Models powering ChatGPT

OpenAI consolidated its separate “o-series” reasoning models into the unified GPT-5 line during 2025–2026, so you no longer pick between a chat model and a reasoning model — the model reasons when needed. Here’s what’s current.

GPT-6 Sol and GPT-6 Luna (regular chat)

Regular chat moved to GPT-6 on 7 October 2026. GPT-6 Sol powers it on Plus, Pro, Business and Enterprise, and GPT-6 Luna on Free and Go from 8 October (OpenAI; MacRumors). The same update added Intelligent UI: answers can include graphics, tappable buttons, forms, charts, diagrams, maps and small interactive tools, such as a bill splitter or a recipe that rescales to the number of guests. OpenAI says that on questions needing a web search, GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant. Both models were already in ChatGPT Work and Codex, where they arrived on 22 September.

GPT-6.1 Sol

GPT-6.1 Sol arrived on 29 September 2026 in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, though not in ordinary chat at launch. Artificial Analysis scores it 52 on its Intelligence Index, one point below GPT-6 Astra, and on the API it costs a fifth of Astra’s price.

GPT-6 Astra

GPT-6 Astra is the current flagship, on every paid tier (Plus, Pro, Business and Enterprise) since 5 September 2026, and Pro’s top reasoning mode runs on it (OpenAI). It scores 53 on the re-based Artificial Analysis Intelligence Index v4.3, level with Claude Fable 5.1 and five points below Claude Opus 5.5, and 62 on the Coding Agent Index. It is the first OpenAI model rated Critical for cyber risk, with that capability gated, and OpenAI published no SWE-bench figure for it.

GPT-5.6 Sol

The flagship on Plus, Pro, Business and Enterprise from 9 July 2026 until GPT-6 Astra reached every paid tier on 5 September. At launch Sol was #2 on the independent Artificial Analysis Intelligence Index and #1 on its Coding Agent Index, and it topped OpenAI’s Terminal-Bench harness; on the re-based v4.3 index it reads 47, six behind Astra — with a new max reasoning effort and an ultra subagent mode. Two caveats: it trails Claude Opus 4.8 on repo-scale SWE-bench Pro (64.6% vs 69.2%), and METR flagged its reward-hacking (“cheating”) rate as the highest of any public model it has tested, so verify autonomous output. On the API, Sol ($5/$30) ships alongside cheaper Terra ($2/$12) and Luna ($0.20/$1.20) tiers, both cut on 30 July 2026.

GPT-5.6 Sol Pro (max effort)

Sol run at its highest max reasoning effort, for the hardest problems — the successor to the old GPT-5 Pro, available on the Pro and Enterprise tiers. It spends more compute for deeper, more reliable reasoning.

GPT-5.5

The prior flagship, released 23 April 2026 and superseded by GPT-5.6 Sol on 9 July 2026 (still selectable on paid tiers). OpenAI framed it less as a smarter chatbot and more as an agent: built to take multi-step actions, hold task state across a session and chain tool calls. It posts about 82.6% on SWE-bench Verified (independently measured), with an ~1.05M-token context window via the API.

GPT-5.6 Luna (the free/Go default until October 2026)

GPT-5.6 Luna was the default on the Free and Go tiers from the week of 6 August 2026 until GPT-6 Luna replaced it on 8 October 2026. It had replaced GPT-5.5 Instant. From the week after it arrived both tiers also got unlimited text chats, subject to abuse guardrails, and a Think button for harder questions; limits still apply to uploads, images, voice and other tools (OpenAI). Paid tiers (Plus and above) get the GPT-6 Astra flagship instead, alongside the GPT-5.6 reasoning models.

GPT-5.4 and GPT-5.3

GPT-5.4 is strong on knowledge-work benchmarks, and GPT-5.3 (with a coding-focused GPT-5.3-Codex variant) remains available on the paid tiers. GPT-5.2 was retired from most tiers in June 2026.

Open-weight models

OpenAI also ships gpt-oss-120b and gpt-oss-20b under an Apache 2.0 licence — its open-weight models for self-hosting, separate from the ChatGPT app.

Retirements

OpenAI retires older models on a rolling basis after a sunset period. o3 left ChatGPT on 26 August 2026 at the end of a 90-day sunset, following GPT-5.1 (11 March 2026) and GPT-5.2 (12 June 2026). Custom GPTs themselves are scheduled for retirement on 11 December 2026 (see Custom GPTs).

Features deep dive

Dots — always-on agents

Announced at DevDay on 29 September 2026, a dot is a persistent agent that runs on GPT-6 Astra with its own cloud computer and browser, works in the background and connects to more than 4,000 apps. You reach it in ChatGPT, on a voice call, or in Slack and Teams, and OpenAI says texting will follow. Every Pro plan includes one, as does a Premium seat on Business, and Enterprise has them in beta. It is not available in the European Economic Area, Switzerland or the UK (OpenAI).

Space and Pages

Also announced at DevDay, Space gives teams and their agents a shared place for files and project context, and Pages are editable documents that teammates and agents can work on together. OpenAI said collaborative slides would follow “in the next few weeks”. ChatGPT also answers @ChatGPT mentions in Slack and Microsoft Teams, using the tools connected to it (OpenAI).

Agent Mode

Agent Mode lets ChatGPT carry out multi-step tasks on your behalf — browsing the web, filling forms, working across tools and holding task state across a session — rather than just answering. OpenAI explicitly frames GPT-5.5 as “an agent, not just a smarter chatbot,” and Agent Mode is available on Plus, Pro, Business and Enterprise.

Codex (agentic coding)

Codex is OpenAI’s agentic coding tool — it reads and writes local files, runs shell commands, connects to GitHub for pull requests and handles multi-step refactors, across the terminal, IDEs and the cloud. It is bundled into the paid ChatGPT tiers and became OpenAI’s fastest-growing product after its May 2026 expansion.

ChatGPT Sites — describe an app, get a hosted URL

ChatGPT Sites, announced at OpenAI’s “Intelligence at Work” event on 2 June 2026 and launched in public beta on 9 July 2026, turns a plain-English description into a working web app that OpenAI builds, hosts and gives you a shareable link to. Powered by Codex, it targets interactive internal tools — dashboards, project trackers, analytics portals, review workspaces and knowledge bases — rather than static marketing pages, and locks access to your workspace through “Sign in with ChatGPT” authentication.

• Prompt-to-app — Describe the tool you want; Codex builds the full application, deploys it to OpenAI-managed hosting and returns a URL
• Workspace-gated — Access is limited to workspace members via Sign in with ChatGPT
• Paid workplace tiers, in beta — Rolling out on paid business/enterprise plans (Free and Go excluded); OpenAI describes it as a preview with its own Sites terms, and it prohibits payment-card data, protected health information and content aimed at under-13s
• Not everywhere yet — Reported as unavailable in the UK, EEA and Switzerland at launch

Web browsing and ChatGPT Search

ChatGPT Search became available to all users (including logged-out visitors) on 5 February 2025. At launch it was built on a fine-tuned GPT-4o with Bing API integration and news partnerships (AP, Reuters, Financial Times). It provides:

• Inline citations with hover-to-preview functionality
• Rich visual results for weather, stocks, sports, maps
• Automatic triggering when queries benefit from current information
• Source transparency — See exactly where information comes from

Deep Research

Deep Research operates as an agentic feature that independently discovers, reasons about, and consolidates insights across the web. Unlike simple browsing, it:

• Autonomously navigates dozens of sources
• Cross-references and fact-checks information
• Produces comprehensive reports with citations
• Takes 5–30 minutes depending on complexity

Monthly limits: Pro gets 250 queries, Plus/Business/Enterprise get 25, Free gets 5.

A July 2025 update added visual browser capabilities—the agent can click, scroll, and navigate websites, pausing at login pages for user confirmation.

GPT-Live — full-duplex real-time voice

On 8 July 2026, GPT-Live replaced Advanced Voice Mode as ChatGPT’s voice experience. The difference is architectural: every previous ChatGPT voice system worked like a walkie-talkie — you talk, it talks, one side at a time. GPT-Live works like a phone call, with both sides live simultaneously. It can listen and speak at the same time, and decides — multiple times per second — whether to speak, keep listening, pause, interrupt or fire a tool call.

• Full-duplex conversation — Listens while it talks, so you can interject naturally and it responds without the old turn-based lag
• Backchanneling — Signals it’s paying attention with acknowledgements like “mhmm” mid-sentence, the way people do
• Search without stopping — Can run a web search while continuing to talk; previously, voice mode paused the conversation whenever it went to the internet
• Two models — GPT-Live-1 serves Plus and Pro; GPT-Live-1 mini serves Free and Go
• Daily limits — since 9 September 2026, three hours on Go and Plus and 15 hours on the $100 Pro plan, with no limit on the $200 plan (OpenAI)

Availability: Rolled out globally from 8 July 2026 across iOS, Android and chatgpt.com. At launch GPT-Live did not support voice with video or screen sharing, which Advanced Voice Mode had. Since 23 September 2026 voice also works with plugins and in Work mode (OpenAI).

Image generation

ChatGPT’s image generation runs on ChatGPT Images 2.5, which rolled out to every ChatGPT tier including Free, plus ChatGPT Work and Codex, on 8 September 2026 (9to5Mac). At launch its two API variants held #1 and #2 on LMArena’s image-editing board on preliminary votes, while its predecessor GPT Image 2 led the Artificial Analysis image arena, which had not yet scored 2.5. DALL·E is fully retired as ChatGPT’s engine — the lineage ran DALL·E 3, then GPT-4o native generation (“GPT Image 1”) from 25 March 2025, then GPT Image 1.5, GPT Image 2 (April 2026) and now Images 2.5.

The model reasons before it renders:

• Instruction faithfulness — accurate multi-element scenes, and up to eight coherent images from a single prompt
• Text rendering — roughly 99% text accuracy across scripts (best AI image generator), the capability DALL·E was worst at
• Resolution — up to 4K
• Conversational editing — “make the sky orange, add a cup on the table, now widen it”
• Commercial rights on every tier, including Free

What Images 2.5 changed (OpenAI’s claims, 8 September 2026): sharper detail, more natural lighting and richer textures; better preservation of the subjects in reference photos; edits that touch only what you asked for and hold across multiple turns; and up to 50% lower latency than Images 2. The update also adds Sketch (draw a rough reference directly in ChatGPT), templates for common formats, and prompt sharing. In the API, Images 2.5 ships as two models at GPT Image 2’s token prices ($5 text input, $8 image input, $30 image output per million): gpt-image-2.5-flare, the fast default, and gpt-image-2.5-sunburst for precision editing, both with new xhigh and max quality levels (model card).

Provenance: outputs carry C2PA metadata and, since May 2026, a SynthID watermark.

Limits and availability: Free gets roughly 2–3 images a day; Plus is capped at about 50 images per 3 hours; Pro is effectively uncapped. Generation on Images 2 took ~30–60 seconds, and OpenAI claims 2.5 cuts that by up to half. Individual users get no IP indemnity: OpenAI’s Copyright Shield covers only Enterprise and API customers. Full comparison in our best AI image generator guide.

Code execution (Advanced Data Analysis)

The code interpreter runs Python in a sandboxed environment with 300+ pre-installed libraries including pandas, numpy, matplotlib, scikit-learn, scipy, and more.

Capabilities:

• Data analysis and visualisation
• File format conversion
• Statistical analysis
• Chart and graph generation
• PDF manipulation
• Image processing

Limitations:

• 120-second runtime limit per execution
• No internet access from the sandbox
• State clears when the environment dies
• Some libraries unavailable for security reasons

File uploads and handling

ChatGPT supports extensive file upload capabilities:

Supported formats: PDF, DOCX, XLSX, CSV, TXT, PNG, JPEG, GIF, WebP, code files (Python, JavaScript, etc.)

Size limits:

• Documents: Up to 512MB per file
• Images: Up to 20MB
• Text content: Up to 2 million tokens (~500,000 words)

Upload limits:

• Plus users: 80 files every 3 hours
• Free users: 3 files daily

Storage caps:

• Individual: 10GB
• Organisation: 100GB

Memory and personalisation

ChatGPT’s memory system has two components:

1. Saved Memories — Explicitly requested or automatically captured details about you
2. Reference Chat History — Rolled out April 2025 for Plus/Pro, June 2025 for Free

This enables ChatGPT to remember your preferences, communication style, interests, and context across conversations. Memory works on every tier, including Free.

Managing memory:

• View: Ask “What do you remember about me?”
• Control: Settings > Personalization > Manage Memories
• Delete: Remove individual memories or clear all
• Notifications: You’re informed when memories are added

Temporary Chat provides incognito mode—conversations aren’t saved to history, aren’t used for training, and don’t update memory. Data deleted within 30 days.

Custom GPTs and the GPT Store

Over 3 million Custom GPTs have been created through OpenAI’s no-code builder. To create one:

1. Describe the desired functionality in natural language
2. Upload knowledge files (up to 20)
3. Configure custom actions for API integration
4. Set conversation starters and personality

Custom GPTs are being retired. OpenAI announced on 11 September 2026 that it will retire custom GPTs and replace them with plugins, which package reusable instructions, reference material and connections to outside services and work in both ChatGPT and Codex. Retirement is scheduled for 11 December 2026, though OpenAI says timing can vary by account and workspace. Existing GPT conversations do not migrate, and custom actions may have to be rebuilt (OpenAI; Virtualization Review).

The GPT Store marketplace serves Plus, Business, and Enterprise users with featured and trending categories across productivity, writing, research, programming, education, and lifestyle.

Monetisation: A revenue-sharing program based on engagement metrics exists for US-based creators, though implementation details remain sparse. Most creators monetise through affiliate marketing, external paywalls, or consulting services.

Canvas for collaborative editing

Canvas provides a dedicated two-pane workspace for writing and coding projects. The feature auto-triggers when ChatGPT detects content exceeding 10 lines or editing scenarios.

Writing shortcuts:

• Adjust length (shorter/longer)
• Change reading level (from 5th grade to graduate)
• Add final polish
• Add emojis
• Translate

Coding shortcuts:

• Review code
• Add logging/comments
• Fix bugs
• Port to other languages (JavaScript, Python, Java, C++, PHP, TypeScript)
• Run code

Key capabilities:

• Inline highlighting for targeted edits
• Version history with “show changes” highlighting
• Export to PDF, Markdown, Word, or language-specific code files

Available on web, Windows and macOS.

Projects for organised workspaces

Projects group related chats and files into dedicated workspaces with persistent context. Originally launched December 2024 for paid tiers, expanded to Free users in 2025 with limitations.

Each project supports:

• Custom instructions specific to that project
• Up to 20 file uploads (40 for Pro users)
• Memory that persists across all project chats
• Colour coding and organisation

Shared Projects (October 2025) allow multiple people to participate—members can start chats and view others’ sessions.

Connectors

Connectors provide direct integration with external tools:

• Google: Gmail, Calendar, Drive
• Microsoft: Outlook, Teams, SharePoint, OneDrive
• Development: GitHub
• Storage: Dropbox, Box
• CRM: HubSpot

This allows ChatGPT to search your emails, check your calendar, find documents, and take actions—all through natural conversation.

Note: Connectors are unavailable in EEA, Switzerland, and UK due to regulatory constraints.

Pro users can build custom connectors via MCP (Model Context Protocol).

Tasks and scheduling

Tasks (launched January 2025) enable one-time or recurring prompts:

• Daily news briefings on topics you care about
• Stock price updates at market open
• Language learning exercises each morning
• Weather reports before your commute
• Weekly report summaries

You can maintain up to 10 active tasks, receiving push and email notifications upon completion.

Finances and shopping

Finances lets you connect financial accounts and ask about spending, saving and investments, with answers grounded in your own data. It reached Free and Go users in the US on 2 October 2026, on web, iOS and Android, and since 21 September Plus and Pro users in the US can also track their credit score, supplied by Experian on the VantageScore 3.0 model. On 1 October 2026 product listings for clothes and accessories gained a Try on button, which generates a virtual try-on from a selfie with ChatGPT Images (OpenAI).

ChatGPT for Teens and ChatGPT for Teachers

ChatGPT for Teens launched on 18 August 2026. Users estimated or stated to be 13 to 17 are placed into a guardrailed version by default, with Study Mode, homework prompts that redirect assignment-shortcutting toward tutoring, and Study Hours that a parent can set. OpenAI frames it as supporting learning “outside of the classroom”.

ChatGPT for Teachers is a separate, free workspace for verified US K-12 educators and school staff, launched 19 November 2025 and verified through SheerID. It includes unlimited messages with the flagship model, connectors into Google Drive and Microsoft 365, shared projects for co-planning, and admin controls for districts — and workspace content is not used for training by default. It is explicitly not for students, and it is US-only: there is no equivalent free tier for teachers elsewhere.

OpenAI’s product page says the plan is free “through June 2028” while its pricing page and the original launch post say June 2027. See our best AI tools for teachers guide for the full comparison.

Video generation with Sora

Sora is OpenAI’s text-to-video model, available inside ChatGPT on paid tiers (with the most capacity on Pro). The standalone Sora app was shut down in March 2026 over unsustainable compute cost (reportedly around $1M/day at peak), but Sora 2 continues with stronger physics, audio and control. Video generation remains usage-capped and carries the usual deepfake and likeness concerns.

Strengths

Reach — More than 900 million weekly users, over three million custom GPTs and the broadest set of third-party integrations of any assistant.

Breadth — Text, images, voice, video, code, research, browsing and agents in one app.

Persistent memory — Memory works on every tier, including Free; Claude limits its memory to paid plans.

Voice — GPT-Live is full-duplex: it listens while it talks and can search the web without pausing the conversation.

Agents — Agent Mode, Codex and dots take on multi-step and ongoing work.

Enterprise controls — SOC 2 and ISO certifications, data residency options and customer-managed encryption keys.

Limitations

Ads on the free tier — The free and Go tiers show ads in a growing number of markets, including the US, UK, India and much of Europe.

Coding isn’t best-in-class on every measure — GPT-6 Astra scores 62 on the AA Coding Agent Index, level with Claude Fable 5.1, but OpenAI published no SWE-bench figure for it, and on SWE-bench Pro Claude’s Fable 5 (80%) and Opus 5 (79.2%) lead GPT-5.6 Sol (64.6%). METR also flagged Sol’s reward-hacking rate as the highest of any public model tested, so verify autonomous output.

Rate limits and cost — Even on Plus you can hit limits during heavy use, and Pro runs from $100 to $500 a month.

Privacy and legal overhang — Training uses consumer data by default (opt-out), and a court ordered OpenAI to retain and produce millions of anonymised ChatGPT logs in the New York Times copyright case.

Ecosystem lock-in — The breadth that makes ChatGPT convenient also concentrates your data and workflows in one vendor.

Connector restrictions — EEA/UK users can’t access many integration features due to regulation.

Video generation limits — Sora is limited to paid tiers and constrained by usage caps.

Who should use ChatGPT?

Great for:

• General AI users — one assistant covering chat, images, video, voice, coding and research
• Visual creators — Images 2.5 reasons before it renders and handles text in images
• Voice-first users — GPT-Live is full-duplex, so you can interrupt it mid-sentence
• Builders — Custom GPTs let you create specialised tools without coding, until plugins replace them in December 2026
• Researchers — Deep Research and web search with citations
• Enterprise teams — Mature security, compliance, and admin features

Look elsewhere if:

• Maximum context is essential — Gemini offers 1M tokens in the app against roughly 400K on ChatGPT’s top tier
• Deep coding is your primary use — Claude models post the highest published SWE-bench Pro scores
• You want a free tier without ads — Claude and Gemini show none
• You live in Google Workspace — Gemini is built into Gmail, Docs and Drive

Historical timeline

Privacy and data handling

Training data policy

To opt out: Settings > Data Controls > “Improve the model for everyone” → Off

Temporary Chat

Temporary Chat provides incognito mode:

• Conversations aren’t saved to your history
• Not used for model training
• Don’t update your memory
• Retained for 30 days for safety monitoring only, then deleted

Data retention

• Regular chats: Retained until you delete them
• Deleted content: Removed from systems within 30 days
• Temporary chats: 30 days maximum, safety monitoring only

Security certifications

Enterprise and Business products hold:

• SOC 2 Type 2 — Security, availability, processing integrity
• ISO 27001 — Information security management
• ISO 27017 — Cloud security
• ISO 27018 — Protection of personal data in cloud
• ISO 27701 — Privacy information management

Visit OpenAI’s Trust Portal for audit reports and compliance documentation.

Data residency

Enterprise customers can choose data residency in: US, Europe, UK, Japan, Canada, South Korea, Singapore, Australia, India, UAE.

Enterprise Key Management (EKM)

Enterprise tier supports customer-controlled encryption keys—your data is encrypted with keys you control, providing an additional security layer.

ChatGPT vs alternatives

Quick comparison

Where ChatGPT wins

• Reach and ecosystem — A billion-plus users, custom GPTs and the broadest integrations
• All-in-one breadth — Text, image, video, voice, agents and research in one place
• Voice — Full-duplex GPT-Live
• Agents — Agent Mode, Codex and always-on dots

Where competitors win

• Claude — Highest published scores on the hardest coding benchmarks, the top score on the independent intelligence index, and no ads
• Gemini — Google Workspace integration, a 1M-token in-app context and Veo video

Getting started

Web

1. Go to chatgpt.com
2. Sign up with email, Google, Microsoft, or Apple
3. Start chatting—no setup required

Mobile

• iOS: Download from App Store
• Android: Download from Google Play

Desktop

Download from openai.com/chatgpt/download:

• macOS: Requires Apple Silicon, macOS 14+
• Windows: Windows 10 or 11
• Linux: In public preview since August 2026 — Ubuntu 24.04 and 26.04 LTS, Debian 13, and Fedora 43 and 44.

The macOS app gained an Apple Messages plug-in on 20 August 2026: ChatGPT can search message history, summarise threads, draft replies and send texts (iMessage, SMS and RCS) through Messages itself, on Apple Silicon only. Sending is gated behind per-message approval by default; OpenAI warns against enabling persistent approval, which removes the final review before a message goes out under your name. You can work in the built-in browser or Chrome and browser actions are supported, but controlling other desktop applications is not yet available on Linux.

Tips for new users

Be specific: “Write a professional email declining a meeting due to a schedule conflict” beats “write an email.”

Use follow-ups: ChatGPT maintains context. Ask for revisions, clarifications, or different approaches.

Try voice mode: Tap the headphone icon for hands-free conversation. Great for brainstorming or when you can’t type.

Explore the GPT Store: Thousands of specialised GPTs for writing, coding, research, image generation, and more.

Set up memory: Tell ChatGPT about yourself—job, preferences, communication style—and it’ll remember across conversations.

Frequently asked questions

Is ChatGPT free?

Yes. The free tier runs GPT-6 Luna by default (from 8 October 2026) with unlimited text chats, subject to abuse guardrails, and a Think button for harder questions, plus use of GPT Store bots, web search, basic voice and a small Deep Research allowance; uploads, images and other tools are limited. Since February 2026 the free tier has also shown ads, starting in the US and now in a growing number of markets.

What is the current ChatGPT model?

Regular chat runs on GPT-6 from 7 October 2026: GPT-6 Sol on Plus, Pro, Business and Enterprise, and GPT-6 Luna on Free and Go from 8 October, with unlimited text chats. GPT-6 Astra has been the flagship on every paid tier since 5 September 2026, and Pro’s top reasoning mode runs on it.

Is ChatGPT Plus worth $20/month?

For most regular users, yes. Plus unlocks the GPT-6 Astra flagship, Agent Mode, Deep Research, image and Sora video generation, Codex and custom GPTs. If you only need a bit more than Free, the Go tier ($8/month) is a cheaper step up, though it can show ads.

What about the Pro tiers?

Pro now comes at $100 (5x Plus usage), $200 (10x Plus in Work and Codex) and $500 (25x Plus, plus Astra Ultrafast). All three add Pro reasoning powered by GPT-6 Astra and one always-on dot outside the EEA, Switzerland and the UK. Most people are well served by Plus.

What is Agent Mode?

Agent Mode lets ChatGPT carry out multi-step tasks for you — browsing, filling forms, using tools and holding task state across a session — rather than only answering. It is available on Plus, Pro, Business and Enterprise.

Is my data private?

Free, Go and Plus conversations may be used for training by default — you can opt out in Settings → Data Controls — and the free tier shows ads in a growing number of markets. Business and Enterprise conversations are never used for training. Note that a court order in the New York Times copyright case has required OpenAI to retain and produce millions of anonymised ChatGPT logs.

Can ChatGPT generate images and video?

Yes — and images on the free tier too, at roughly 2–3 a day. Image generation runs on ChatGPT Images 2.5 (from 8 September 2026), with commercial rights granted on every tier; Plus is capped at about 50 images per 3 hours. Sora produces video on paid tiers (with the most capacity on Pro). The standalone Sora app was shut down in March 2026, but Sora 2 continues inside ChatGPT.

What is GPT-Live?

GPT-Live is ChatGPT’s voice experience, launched on 8 July 2026 to replace Advanced Voice Mode. It’s full-duplex — it listens and speaks at the same time, like a phone call rather than a walkie-talkie — so it can be interrupted naturally, backchannel with “mhmm,” and even search the web without pausing the conversation. GPT-Live-1 serves Plus and Pro; GPT-Live-1 mini serves Free and Go. At launch it did not support voice with video or screen sharing.

Does ChatGPT remember previous conversations?

Yes. Memory stores details about you across conversations, which you can view, edit and delete in settings. Use Temporary Chat for sessions that aren’t remembered.

Verdict

ChatGPT has more than a billion monthly users and the broadest feature set of the major assistants: GPT-6 Astra, Agent Mode, Codex and dots for multi-step and ongoing work, plus image generation, Sora video, voice, Deep Research and scheduled tasks.

Free ($0): Unlimited text chats on GPT-6 Luna, with ads in a growing number of markets and tight limits on uploads, images and tools.

Go ($8) / Plus ($20): Go is a cheap step up with more uploads, images and tools, though it can show ads; Plus is the plan for most people, adding the GPT-6 Astra flagship, Agent Mode and the full feature set.

Pro ($100, $200 or $500/month): For heavy professional users who want 5x, 10x or 25x Plus usage, Pro reasoning powered by GPT-6 Astra and an always-on dot; the $500 plan adds Astra Ultrafast.

Claude posts the highest published scores on the hardest coding benchmarks and has no ads, and Gemini is built into Google’s apps. See ChatGPT vs Claude for the head-to-head.

Official links