ChatGPT Review (July 2026): Plans, Features, Privacy, and the Verdict

Quick answer. ChatGPT is OpenAI’s general-purpose AI assistant, available on web, iOS, Android, macOS, and Windows, with tiers running from Free to Pro and Enterprise. As of July 2026 it is powered by the GPT-5.5 Instant and GPT-5.6 model families, ships with browsing, image generation, memory, deep research, scheduled tasks, an agent mode, plugins, and optional ads on Free and Go. Based on its documented feature set and policies, it is a strong general-purpose option for knowledge work and for teams that want one assistant for many jobs — but citation-first research tools and IDE-native coding tools remain useful additions for specialized work.

What is ChatGPT?

ChatGPT is a generative AI chatbot built by OpenAI. It uses large language models (LLMs) from the GPT family to turn your text, voice, or image prompts into replies, images, code, spreadsheets, presentations, and increasingly, actions on the web. OpenAI runs ChatGPT as a freemium service: a free tier, two consumer paid tiers (Plus and Pro), a mid-priced consumer tier called Go, and two workspace tiers called Business and Enterprise. Education, healthcare, academic, and clinician variants sit on top of that base.

You can use ChatGPT in a browser at chatgpt.com, in dedicated apps for iOS, Android, macOS, and Windows, and now through plugins and “Sign in with ChatGPT” on partner sites such as Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel.

How ChatGPT was evaluated in July 2026

This review scores ChatGPT as a daily general-purpose assistant, not as a research citation engine, an IDE coding product, or an enterprise compliance suite. The evaluation is documentation- and source-based: it draws on OpenAI’s pricing page, release notes, privacy policy, enterprise privacy page, data controls help article, and launch posts for Deep Research, ChatGPT Agent, GPT-5.5, and GPT-5.6, with adoption and history cross-checked against the Wikipedia entry, a Stanford HAI AI Index report, a NIST publication, and WIRED reporting. No independent product tests or firsthand benchmarks were conducted; numerical claims below come from OpenAI’s published reports unless otherwise attributed.

“ChatGPT in 2026 is less a single chatbot and more a control plane for a stack of models, tools, and agents — and the gap between Free and Pro is wider than the $200 price tag suggests.”

What changed in ChatGPT over 2025 and into 2026

The most useful starting point for a July 2026 evaluation is what shipped in the preceding eighteen months. This timeline is documented in OpenAI’s release notes and announcement posts.

1. February 2, 2025 — Deep Research launches. A long-running research agent that browses the web and writes a cited report. OpenAI says it can finish in 5 to 30 minutes what would take a human many hours.
2. March 25, 2025 — Native image generation in GPT-4o. Replaces DALL·E 3 as the default image model in ChatGPT; ships with C2PA metadata so images can be identified as AI-generated.
3. May 5, 2025 — GPT-5.5 Instant becomes the default model. Replaces GPT-5.3 Instant for all ChatGPT users; OpenAI says it is more factually reliable and less over-formatted.
4. July 17, 2025 — ChatGPT Agent launches. A unified agent that can browse, click, run code, and deliver editable documents and spreadsheets. Built on Operator and Deep Research, with a “Watch Mode” for high-stakes actions.
5. December 2025 — Ads begin testing in ChatGPT Free for logged-in US adults.
6. January 1, 2026 — Updated Terms of Use go into effect.
7. April 23, 2026 — GPT-5.5 family rolls out to Plus, Pro, Business, and Enterprise in ChatGPT and Codex.
8. April 28, 2026 — Updated Privacy Policy published.
9. May 14, 2026 — Codex remote access from the ChatGPT mobile app becomes available across all plans.
10. June 4, 2026 — Lockdown Mode is enabled for all signed-in users. A strict, opt-in setting that disables web browsing, deep research, agent mode, and file downloads to reduce the risk of prompt-injection data leaks.
11. July 9, 2026 — ChatGPT Work launches for paid plans. An agent for “longer, more involved tasks” that creates finished documents, spreadsheets, presentations, and Sites; replaces the older App Directory with a new Plugin Directory.
12. July 9, 2026 — GPT-5.6 family (Sol, Terra, Luna) reaches general availability. Sol is the new flagship; Luna is the cheapest tier; Terra sits between them.
13. July 23, 2026 — Health in ChatGPT begins rolling out in the US. A separate health workspace that connects supported medical records and Apple Health, with explicit rules that the data is not used to train models or target ads.
14. July 29, 2026 — ChatGPT for Academic Researchers program opens. Verified faculty and postdocs can apply for twelve months of complimentary Pro-level access in a dedicated workspace.

ChatGPT pricing and plans (July 2026)

ChatGPT today is sold in six tiers. The exact consumer prices are listed in US dollars on the official pricing page; Business and Enterprise are quoted per seat and Business starts at two seats. Here is what the plans look like side by side, drawn from OpenAI’s pricing page and release notes.

Plan	Price (US)	Default model	Best for	Notable limits
Free	$0	GPT-5.5 Instant (limited)	Trying ChatGPT, light personal use	Limited messages and uploads; limited deep research; limited memory and context; limited image generation
Go	Low monthly price (Go-tier)	GPT-5.5 Instant (expanded)	Heavy personal users on a budget	More messages, more uploads, more images, longer memory; may include ads
Plus	$20 / month (Plus tier)	GPT-5.5 Instant plus GPT-5.6 Sol reasoning	Solo professionals and creators	Expanded memory, projects, scheduled tasks, custom GPTs, image generation with Thinking
Pro	$200 / month (Pro tier)	GPT-5.6 Sol Pro for highest-stakes work	Power users, researchers, agent-heavy workflows	“5x or 20x more usage,” Pro-only reasoning, maximum Codex tasks, research preview of new features
Business	Per seat, 2+ seats	GPT-5.6 Sol plus workspace admin tools	Small and mid-sized teams	Dedicated workspace, SOC 2 Type 2, data residency options in ten regions, admin console, SSO
Enterprise	Contact sales	GPT-5.6 Sol plus enterprise controls	Larger organizations and regulated industries	SAML SSO, SCIM, custom data retention windows, compliance API, expanded GPT controls, dedicated onboarding

Two fine prints that matter:

• Ads. OpenAI’s pricing page and release notes confirm that the Go plan “may include ads” and that Plus, Pro, Business, Enterprise, and Education remain ad-free. The Wikipedia entry independently confirms the company began testing ads for logged-in US adults in January 2026.
• Student and nonprofit discounts. The same pricing page confirms a free plan for verified US K–12 teachers through June 2027, an Education plan for universities, and up to a 75% discount on Business or Enterprise for qualifying nonprofits.

What can ChatGPT actually do in 2026?

This section separates documented features from marketing language. Each point below is anchored to OpenAI’s product pages and release notes.

1. Chat, reasoning, and writing

The core experience is the same as 2024 — type or speak, get text back — but the engines behind it have changed. OpenAI says GPT-5.5 Instant is more factually reliable and less over-formatted than GPT-5.3 Instant, and that GPT-5.6 Sol “sets a new standard for both intelligence and efficiency” while using fewer tokens. The release notes for GPT-5.5 also document a June 24, 2026 update to GPT-5.5 Instant that improved multi-turn context, complex-instruction following, and shopping and local business answers.

Writing, summarization, and explanation remain core ChatGPT use cases. Confident-but-wrong answers (the so-called hallucination problem) are still possible. OpenAI’s own privacy policy is explicit: “You should not rely on the factual accuracy of output from our models.”

2. Image generation and vision

ChatGPT now generates images natively through GPT-4o’s image model rather than routing prompts to DALL·E 3. According to OpenAI’s launch post, the new model renders in-image text far more accurately, supports multi-turn refinement, and embeds C2PA metadata so downstream tools can flag images as AI-generated. On paid plans you can use image generation with Thinking, which trades speed for higher quality. Vision (uploading an image to ask about it) is available on all paid tiers and “limited” on Free.

3. Deep Research

Deep research is OpenAI’s name for an asynchronous research agent that browses the web, reads sources, and returns a cited report. The original launch post said it takes “anywhere from 5 to 30 minutes to complete its work.” Usage is unlimited on Pro, capped on Plus, and limited on Free and Go. A 2026 update lets you connect deep research to any MCP or app and restrict web searches to a list of trusted sites — useful when you only want primary sources.

4. ChatGPT Agent and ChatGPT Work

ChatGPT agent launched July 17, 2025 as a single experience that combines Operator-style web interaction with Deep Research-style synthesis. It runs in its own virtual computer, can prompt you to take over the browser for logins, and explicitly asks before consequential actions like making a purchase. The launch post assigns it a 41.6 on Humanity’s Last Exam and 68.9% on BrowseComp.

ChatGPT Work, which began rolling out July 9, 2026, is the successor framing. It is an “agent for longer, more involved tasks” that can research, build connected-app workflows, create finished documents, spreadsheets, presentations, and Sites, and run on a schedule. On web and mobile it is rolling out to Pro, Pro Lite, Enterprise, and Edu first, then Plus and Business, and it is off by default during a two-week Enterprise and Edu preview.

5. Voice, video, and the new Voice model

ChatGPT Voice got a new engine on July 8, 2026: GPT-Live-1 for paid users and GPT-Live-1 mini for Free. Both can listen and speak at the same time. Voice with Video is available on Go, Plus, and Pro. A voice dictation update on June 26, 2026 reduced the speech-to-text word error rate by at least 10% on the top languages OpenAI tested.

6. Memory, projects, and custom GPTs

ChatGPT can remember details between chats through two different systems. Saved memories are facts the assistant has decided to remember (your role, your preferred format, your dog’s name). Memory with past chats, rolled out in stages through 2025 and 2026, lets ChatGPT pull context from your recent history. You can turn memory off entirely, delete individual memories, or use a Temporary Chat that is not used for memory or training and is auto-deleted after 30 days. Custom instructions, which sit in a separate setting, can now run up to 5,000 characters on paid plans, up from 1,500.

7. Scheduled tasks, Codex, and developer workflows

Scheduled tasks let ChatGPT run recurring work — a morning briefing, a weekly report, a daily inbox summary — without you opening a chat. Codex is the dedicated coding agent; on July 9, 2026, OpenAI replaced the old Codex app with a new ChatGPT desktop app that bundles Chat, Work, and Codex into one window, and on May 14, 2026 it made Codex remote access from the mobile app available across all plans.

8. Health in ChatGPT and other specialized experiences

Health in ChatGPT, rolling out in the US since July 23, 2026, is a separate workspace for medical records and Apple Health data. OpenAI says the connected data and conversations that use it are not used to train foundation models or target ads. There are similar targeted experiences for clinicians, academic researchers, and teachers.

ChatGPT privacy: what the documentation says

If you only read one section of this review, read this one. OpenAI’s privacy story in 2026 is layered, and the right answer depends on which plan you pick.

• Free and Go. Your content can be used to train OpenAI’s models unless you opt out. The opt-out lives in Settings → Data Controls and is called “Improve the model for everyone.” It syncs across web and mobile. The Go plan may also show ads.
• Plus and Pro (personal). Same opt-out model. Your content can be used to train models by default; you have to flip the switch to stop it.
• Business, Enterprise, Edu, and Healthcare. By default, OpenAI does not train on your content and does not use it to improve services. This is confirmed by both the Enterprise privacy page and the consumer Privacy Policy. Deletions are removed from OpenAI’s systems within 30 days.
• Temporary Chats. Not used for training, not saved in history, and deleted after 30 days. Use these for one-off sensitive asks.
• Lockdown Mode. Rolled out to all signed-in users on June 4, 2026. Turning it on disables web browsing, deep research, agent mode, and file downloads to reduce the risk of data leaks through prompt injection.
• Encryption. OpenAI states on both the pricing page and the Enterprise privacy page that data is encrypted in transit with TLS 1.2 or higher and at rest with AES-256. The security team operates on a 24/7/365 on-call rotation.
• Compliance. The Business, Enterprise, Edu, and Healthcare tiers have each completed a SOC 2 Type 2 audit. You can verify on the OpenAI Trust Portal.
• Your data export and deletion. You can export your chat history and data from the data controls screen, delete individual conversations, delete your account, or use a Temporary Chat for one-off sensitive work.
• Worth saying out loud. OpenAI’s own Terms of Use remind you that you should not treat any single ChatGPT answer as a “sole source of truth or factual information” and that you should evaluate outputs before using or sharing them. The privacy policy goes further: “you should not rely on the factual accuracy of output from our models.”

What ChatGPT is best for

Based on the documented product surface and confirmed features above, ChatGPT is best suited to:

• You need one assistant for mixed work. Drafts, summaries, explanations, brainstorming, light analysis, image generation, scheduled tasks, and research are all in one place.
• You want documented voice and video features on mobile. Voice with Video is documented for Go, Plus, and Pro, while GPT-Live-1 serves paid users and GPT-Live-1 mini serves Free.
• You need an agent you can supervise. ChatGPT Agent and ChatGPT Work include documented Watch Mode prompts and takeover flows for sensitive actions.
• Your team is standardizing on a single AI baseline. Business tier gives you a dedicated workspace, an admin console, SSO, and a data-residency menu that includes the US, EU, UK, Japan, Canada, Korea, Singapore, India, Australia, and the UAE.
• You want images, decks, and spreadsheets generated from chat. Image generation with Thinking, Sites, and the Excel/Google Sheets and PowerPoint plugins make ChatGPT a strong option for non-developers who need finished artifacts.

What ChatGPT is not best for

Based on the documented capabilities and limitations, consider alternatives when:

• Your job is citation-first research. Deep Research cites sources, but OpenAI’s launch post warns that it can hallucinate facts and struggle to distinguish authoritative information from rumors. For academic or medical work, use a primary-database tool alongside ChatGPT.
• You write production code all day. IDE-native tools such as GitHub Copilot, Cursor, and Windsurf are built around repository-level workflows, whereas ChatGPT is a general-purpose assistant with Codex integration.
• You need guaranteed factual accuracy. No LLM gives you that. Cross-check anything that goes to a customer, a patient, or a court.
• You are in a regulated industry and not on Enterprise. Free, Go, Plus, and Pro do not get the SOC 2, SAML SSO, SCIM, or data-residency guarantees that regulated buyers usually need.
• You need a citation-rich web search experience as your default. ChatGPT Search exists, while Perplexity is built around citation-first answers and is an alternative to evaluate if every answer needs a footnote.

ChatGPT alternatives in July 2026

If ChatGPT is not the right fit, these products merit comparison for the use cases below; this review does not rank them through independent testing.

• Claude (Anthropic) for longer-document reasoning and coding workflows where you want a different stylistic and safety profile.
• Gemini (Google) if you live in Google Workspace and want a model that is tightly integrated with Gmail, Docs, and Drive.
• Microsoft Copilot if your team is on Microsoft 365 and wants the assistant built into the apps you already pay for.
• Perplexity for citation-first web search and competitive intelligence where every answer needs a link.
• GitHub Copilot, Cursor, and Windsurf for engineers who need repository-aware coding.
• Notion AI, Granola, and Figma Make for note-taking, meeting notes, and design-to-code, respectively, if your work is concentrated in one of those surfaces.

Pros and cons at a glance

Pros

• A broad single-product surface: chat, voice, image generation, deep research, scheduled tasks, agents, and plugins in one app.
• Mature model lineup with three named tiers (Sol, Terra, Luna) under GPT-5.6, and a separate GPT-5.5 Instant fallback.
• Documented availability across mobile and desktop, with a new unified ChatGPT desktop app rolled out July 9, 2026.
• Strong privacy controls for users who want them: opt-out of training, Temporary Chats, Memory off, Lockdown Mode, and a chat-history export.
• Workspace-grade compliance on Business, Enterprise, Edu, and Healthcare, including SOC 2 Type 2 and 10-region data residency.
• Documented safety controls: Lockdown Mode restricts web-connected features, while Watch Mode adds safeguards around consequential actions.

Cons

• Hallucination risk remains documented. OpenAI itself tells you not to rely on model output as the sole source of truth.
• Content on Free, Go, Plus, and Pro may be used to improve models unless you opt out; the Go plan may also include ads.
• Pro at $200 / month is steep for solo users who do not need Pro-only reasoning or “research preview of new features.”
• Agent features are still early. OpenAI’s own launch post says slideshow output “can sometimes feel rudimentary” and that discrepancies between the viewer and the exported PowerPoint are being reduced.
• Chat history deletion is not immediate. Deleted conversations are removed from systems within 30 days, and Business and Enterprise data is governed by workspace retention windows you have to set yourself.
• Pricing is volatile. Plan names, prices, and limits have shifted at least once a year since 2023. Confirm on the official page before you commit.

FAQ: ChatGPT in 2026

Is ChatGPT worth paying for in 2026?

Based on the documented plan differences, Plus is the clearest paid upgrade for many frequent individual users. It includes expanded memory, longer context, image generation with Thinking, scheduled tasks, custom GPTs, and the GPT-5.6 reasoning models. Pro is suited to users who need higher limits for heavy agent use, large-volume research, or API-adjacent workflows. Free suits casual use and trial but has lower message and image caps.

What is the difference between ChatGPT Free, Plus, and Pro?

Free gives you a limited GPT-5.5 Instant with restricted messages, uploads, deep research, and image generation. Plus ($20 / month at the time of writing) adds GPT-5.6 reasoning, Projects, scheduled tasks, custom GPTs, image generation with Thinking, and expanded memory and context. Pro ($200 / month) layers “5x or 20x more usage,” GPT-5.6 Sol Pro reasoning, maximum Codex tasks, and early access to new features. Confirm current pricing on the official page before you buy.

How much does ChatGPT Pro cost?

As of July 2026, ChatGPT Pro is $200 per month on the official pricing page. The actual price you see at checkout can vary by region, taxes, and any promotional credits; the pricing page lists the canonical US figure.

Can ChatGPT browse the web and cite sources?

Yes. ChatGPT Search is the default web search experience, and Deep Research is the long-form research agent that writes cited reports. Lockdown Mode, if you turn it on, disables both. Deep Research’s own launch post notes that it can sometimes “hallucinate facts” and “struggle with distinguishing authoritative information from rumors” — so even cited answers deserve a second look.

Is ChatGPT safe to use for work and private data?

For personal and knowledge work, the documented controls support turning off model improvement and using Temporary Chat for sensitive one-off requests. For regulated or customer data, use Business, Enterprise, Edu, or Healthcare, where OpenAI confirms it does not train on your content by default and where documented controls include SOC 2 Type 2, SAML SSO, and data residency in ten regions. OpenAI states that its 24/7/365 security on-call rotation and AES-256 / TLS 1.2+ encryption apply to every plan.

Does ChatGPT train on my conversations?

On Free, Go, Plus, and Pro it can, unless you opt out. The opt-out is in Settings → Data Controls → “Improve the model for everyone.” On Business, Enterprise, Edu, and Healthcare, OpenAI confirms it does not train on your content by default. Temporary Chats are never used for training.

What is ChatGPT Memory and how do I turn it off?

Memory is the system that lets ChatGPT remember facts and pull context from past chats. You can turn it off in Settings → Personalization → Memory, view and delete individual memories, and run Temporary Chats that are not used for memory or training. The original Memory post is dated February 13, 2024; the “memory with past chats” upgrade rolled out through 2025 and 2026.

What can ChatGPT Agent actually do?

According to OpenAI’s launch post, ChatGPT Agent can browse the web, click, fill forms, log into sites (with you taking over the browser for sensitive flows), run code, deliver editable documents and spreadsheets, and create slideshows. It scored 41.6 on Humanity’s Last Exam and 68.9% on BrowseComp at launch, and the new ChatGPT Work layer rolling out from July 9, 2026 adds scheduled tasks, Sites, and finished documents and presentations.

How does ChatGPT compare to Claude and Gemini?

The sources used for this review do not support a direct July 2026 performance ranking, and the three products were not independently tested side by side. The 2025 AI Index from Stanford HAI found that “performance differences on major benchmarks such as MMLU and HumanEval shrank from double digits in 2023 to near parity in 2024” between US and Chinese models; that broader finding does not determine which current assistant is best. At a category level, consider Claude when long-form reasoning and writing style are priorities, Gemini when Google Workspace integration matters, and ChatGPT when feature breadth and agent tools matter.

Who should use ChatGPT Enterprise?

Based on its documented controls, ChatGPT Enterprise is aimed at organizations in regulated industries, larger deployments, and teams whose security requirements include SSO, SCIM, audit logs, custom data retention, or data residency. Enterprise also documents access to custom models, granular GPT controls, and dedicated workspace administration. For a small startup without compliance pressure, the documented Business features may be sufficient.

Verdict: source-based editorial take

This July 2026 score is based on the cited documentation and reporting, not firsthand product testing. The unified desktop app, the GPT-5.6 lineup, the Work agent, Lockdown Mode, and the updated privacy policy together document a single assistant designed to cover writing, research, image generation, voice, and supervised automation for a broad audience.

The documented limitations that keep the score below 90 are not new. Hallucination is still a real risk that OpenAI itself documents. On Free, Go, Plus, and Pro, OpenAI may use content to improve models unless users opt out. Agent features have documented limitations — OpenAI warns that Deep Research can hallucinate and says slideshow output can sometimes feel rudimentary, while Lockdown Mode addresses prompt-injection risk. And the $200 / month Pro price is only justified when its higher limits and Pro-only features match the buyer’s needs.

The bottom line: based on its documented feature breadth, ChatGPT is a strong general-purpose AI default in 2026. Use it for writing, analysis, research, and supervised automation; add a citation-first tool for high-stakes research and an IDE-native coding tool for sustained software development. Important facts still need verification, as OpenAI’s own terms and privacy policy make clear.

Final score: 88 / 100. Strong documented versatility, a mature product surface, and a broad agent feature set — held back by documented hallucination risk and by a privacy posture on Free, Go, Plus, and Pro that requires users to review their data-control settings.