ChatGPT vs Claude: Similarities and Differences (plus alternatives)

So, you’re getting deeper into using AI tools. But before you go much further, you want to know which company you should be getting into bed with.

Much like Android and iOS, there are loads of things in common, but a few distinguishing differences. Switching teams can be time-consuming as you learn the ins and outs of the other, so it is useful to pick a side and stick with it for a while.

Below is an exhaustive (I hope) analysis of what ChatGPT and Claude — and their companies, OpenAI and Anthropic — have that distinguishes each one, what they have in common, and whether there are realistic alternatives.

I use both with paid plans and switch their priority every now and then, so I am personally aware of their pros and cons. It is impossible to cover everything, but this should be enough for most everyday users.

The short version

Below is a summary table of why choose ChatGPT or Claude.

If I had to recommend one $20 plan to a general user, it would be ChatGPT Plus. It covers more jobs without another subscription, especially if you make images or use live voice.

If most of your work is writing, reading long files, building small tools or working through a codebase, Claude Pro can make more sense.

(Note, I don’t get any commission for this post… nobody would!)

What ChatGPT and Claude have in common

In a normal chat window, the products are much more alike than their fans sometimes admit. Both can:

• Answer questions, edit writing, translate and brainstorm.
• Reason for longer before answering when you select a thinking mode.
• Search the web and cite sources.
• Run a longer research job across the web, files and connected services.
• Read PDFs, spreadsheets, screenshots, photographs, diagrams and charts.
• Write and run code, analyse data and create downloadable files.
• Remember information across chats, subject to your settings and plan.
• Organise work into projects with instructions and reference files.
• Talk by voice on supported devices.
• Connect to third-party services and use tools rather than only producing text.
• Hand longer jobs to an agent that can work through several steps.
• Provide developer APIs, team plans and enterprise controls.

Both also make mistakes. A polished answer is not proof that the answer is correct. Web citations help, but a model can cite a source that does not quite support its claim. For legal, medical, financial or business-critical work, read the original source.

The current models

As of August 2026, OpenAI’s current family is GPT-5.6. Sol is the flagship, Terra is the lower-cost workhorse and Luna is the fastest. ChatGPT Free and Go use Luna by default; paid plans can reach Sol in the main product, but your available usage will run out more quickly. Codex and Work expose a different mix depending on the plan.

Anthropic’s range needs a little more explanation. Claude Fable 5 is its most capable generally available model. Anthropic describes it as a Mythos-level model for ambitious coding, research and agent jobs that may run for days. It is available in Claude, Claude Code, Cowork and the API, but it is not simply bundled into every $20 subscription: on Pro and standard Team seats it uses separately purchased usage credits; Max and premium Team seats can use Fable for up to half of their shared weekly allowance before credits take over.

Claude Opus 5 sits just below Fable at a lower cost and is intended for hard everyday reasoning and agent work. Claude Sonnet 5 is the faster workhorse. In normal use, most people will spend much more time with Sonnet or Opus than Fable, but leaving Fable out gives the wrong picture of Anthropic’s current ceiling.

I would not choose a subscription from one benchmark chart. The result changes with the task, prompt, tool access and model setting. New model releases also swap the lead regularly. The product around the model — limits, files, memory, apps and agent controls — is the difference you will notice every week.

Everyday chat, speed and tone

Speed

There is no permanent winner on speed. A fast model such as GPT-5.6 Luna will usually answer before a flagship model that is allowed to think. Claude Haiku, when offered for a task, serves the same fast-model role on Anthropic’s side. Sonnet and Terra sit nearer the middle; Opus and Sol at high reasoning settings can take longer.

The job matters more than the logo. Ordinary chat normally takes seconds. A deep research report, a large coding change or a computer-use task can take minutes. Server load, tool calls and a site’s own response time can dominate the model’s speed.

Tone

A widespread observation is that Claude’s default prose is often measured, complete and slightly formal. ChatGPT’s default can feel more conversational and can vary more when you change its personality, memory or instructions. That is an observation about defaults, not a testable rule about which model “writes better”.

Both will follow a useful style brief. Give either one a sample, say who the reader is, ban phrases you dislike and ask it to edit its first draft. The gap gets much smaller. Tone also changes after model updates, so I would not make it the sole reason to lock myself into one company.

Web search and deep research

Both products have quick web search and a slower research mode. Quick search is for a current fact. Deep research is for “read these sources, compare them and produce a cited report”.

ChatGPT Deep Research can search the public web, uploaded files and connected apps. You can restrict it to chosen sites, edit the proposed research plan and steer it while it works. App access inside Deep Research is read-only.

Claude Research searches the web and can include connected sources such as Google Workspace. It returns a cited answer after working through the question for several minutes.

The two are close enough that source quality and your instructions matter more than the badge. Tell the agent which sources count, ask it to separate fact from inference, and open the citations. Research allowances are plan-dependent on both services and are lower than ordinary chat allowances.

Images: recognition, analysis and generation

Reading images

Both can inspect photographs, screenshots, scanned pages, graphs and diagrams. They can extract text, describe a scene, explain an interface, compare two images and reason about visual information. Both can also miss small text, count badly, or infer something that is not visible. Crop the relevant area and provide the original file rather than a compressed screenshot when accuracy matters.

Making images

This is a real difference. ChatGPT has native image generation and editing. It can create illustrations and photographs, change a supplied image, preserve transparency and iterate in the same conversation. OpenAI’s capability overview includes image creation as a standard ChatGPT tool.

Claude does not generate ordinary raster images. It can write SVG, HTML and code for diagrams, charts and interactive graphics. Claude Design and Artifacts can make layouts, prototypes, slides and one-pagers, but that is not a substitute for a native photo or illustration generator.

If you make blog covers, product mock-ups or social images, ChatGPT removes the need for a second service. I have also written about a dedicated AI cover-image workflow for WordPress if that is the main job you need to automate.

Files, data analysis and working documents

ChatGPT can run Python in a secure environment for calculations, charts and file transformations. Work can also produce documents, spreadsheets, slides, reports and small sites. Its Canvas gives longer writing and code a dedicated editing area.

Claude’s Artifacts put the output in a visible side panel. An Artifact may be a document, web page, diagram, component or small interactive tool, and it can be revised without burying the result in the conversation. Cowork can create and edit Office files and other local files.

The underlying abilities overlap. The difference is how the work is presented. Claude gives Artifacts a very prominent role. ChatGPT spreads comparable work across Canvas, Work, file tools and Sites. If you spend all day revising one document or small tool beside a chat, Claude’s layout can feel tidier. If you move between several output types, ChatGPT’s wider tool set can save a hand-off.

Computer use: Claude Cowork vs ChatGPT Work

Computer use means the model can do more than tell you which buttons to press. It can browse, click, type, run code and work through a multi-step task. Both companies now put this inside a general work agent.

OpenAI describes Chat, Work and Codex as three different modes: conversation, general multi-step work and software development. Anthropic describes Cowork as a general agent that can work through a task using files, apps, code and the web.

Neither should be given unlimited access without thought. Use a restricted folder, review destructive actions, and keep backups. Anthropic says Cowork tries connectors before visual clicking when it can because an app’s structured interface is faster and less error-prone. The same principle applies to any computer agent: a proper tool or API is usually safer than moving a mouse by sight.

Coding: Claude Code vs Codex

The products are now much more similar than a simple “Claude has a terminal tool; OpenAI has a web agent” comparison suggests. Both can inspect a repository, edit several files, run commands and tests, explain their changes, review code and hand longer jobs to an isolated cloud environment.

Where Codex is stronger

Codex has particularly broad distribution. The same product spans the terminal, a native VS Code-style extension, Xcode, JetBrains, the ChatGPT desktop app, the web, GitHub-connected cloud jobs and iOS. Its cloud mode is built around isolated environments, parallel attempts and reviewable diffs. At the model level, GPT-5.6 Sol is especially strong on command-line and tool-heavy evaluations, and OpenAI’s multi-agent Ultra mode can trade more compute for a higher completion rate.

The catches are that Work and Codex share a usage budget, cloud environments need to be configured correctly, and the Codex IDE extension does not currently support ChatGPT plugins. A high-effort or multi-agent run also consumes more resources than a quick local edit.

Where Claude Code is stronger

Claude Code has an established terminal-centred workflow, a good native VS Code interface for reviewing plans and inline diffs, and a close relationship with Cowork, Claude’s skills and its plugin system. It can also work with Claude in Chrome for visible browser testing. Fable 5 gives it a high-capability option for large migrations, complex implementations and long autonomous jobs; Anthropic says Fable writes tests and uses vision to compare rendered work with the goal.

The catches are practical. Claude chat, Code and Cowork draw from the same subscription allowance. Fable is expensive: Pro users pay for it through usage credits, while Max includes only a defined portion inside the weekly allowance. Fable also has extra safety routing in sensitive biology and cybersecurity areas, which can switch a request to a lower model.

What the current measurements show

No single coding score settles this. Benchmarks test different jobs and depend on the agent harness, tools, reasoning budget and time allowed. The useful result is that the lead changes with the test.

Those figures are the head-to-head results reported in OpenAI’s GPT-5.6 release table; the Artificial Analysis index itself is independent. OpenAI also reports that Sol reached its index score using less than half Fable’s output tokens, in less than half the time and at about one-third of the estimated cost. Fable’s large SWE-Bench Pro lead points the other way. The published table does not yet give a comparable Opus 5 result, so I would not invent one.

In plain English: current measurements support Codex/Sol for fast, tool-heavy implementation and terminal jobs, while Fable has the strongest published result here on difficult repository issues and is designed for very long-running work. On a normal paid plan, the more realistic comparison may be GPT-5.6 Terra or Sol against Claude Sonnet 5 or Opus 5, because Fable can incur extra charges.

For developers, model access through an API is a separate purchase from the consumer subscription. A ChatGPT Plus or Claude Pro subscription does not give you a pot of API tokens. If you are building an integration, structured output matters as much as chat quality; I have a practical example using Zod with OpenAI structured outputs.

Plugins, apps, connectors and tools

The names are confusing because both companies have changed them. In ChatGPT, the old App directory moved into a Plugin directory in July 2026. A plugin can contain skills, apps and templates. Apps connect data or actions from other services. The same plugin system can serve ChatGPT Work and Codex.

Claude also has a directory for skills, connectors and plugins. A plugin may bundle instructions, connectors and subagents for a particular job. Connectors can use remote services, while desktop extensions can reach approved local tools.

Both can use MCP, the open protocol first released by Anthropic for connecting models to tools and data, and which surprisingly became mainstream (I really thought it would just ne an Anthrpic thing!). When investigating plugins, check whether the service you actually use — Google Drive, Slack, GitHub, your CRM or your database — is supported on your plan and whether it can only read or can also make changes.

Connected tools turn a chatbot into something closer to the business assistant described in my plain-English guide to running a business with AI and MCP. They also increase the damage a bad instruction could do. Give each connection the least access it needs and keep approval on for consequential actions.

Other useful differences

• Voice: Both support voice. ChatGPT has a broader live voice experience, including camera/screen input on supported devices and voice coordination with Work and Codex.
• Projects and memory: Both can keep project files and instructions and remember information across chats. Controls, availability and exact limits vary by plan.
• Custom assistants: ChatGPT has custom GPTs and a public directory. Claude leans more on Projects, styles, skills and plugins.
• Scheduled work: Both can schedule supported agent tasks. ChatGPT Work also supports conditional jobs that run when a stated condition becomes true.
• Recording and study: ChatGPT has dedicated recording and study modes. Claude can transcribe, teach and work from uploaded material, but the product packaging differs.
• Context: Both can handle long documents. Published context limits vary by model and product surface; a large advertised window does not mean every plan, tool and conversation receives it.

Usage limits in real life

This is one of the most important differences, and the hardest to reduce to a number. Both companies adjust limits according to the plan, model, tool and demand. The product normally shows a counter or reset time when a specific allowance matters.

How often do people actually hit them? Neither company publishes a reliable percentage, so claims that “nobody” or “everybody” reaches a cap are guesswork. In normal short chats, many paid users will not notice. A long coding session, a folder of documents, repeated Opus use or an afternoon in Cowork can make Claude Pro’s shared allowance visible. ChatGPT users are more likely to meet a limit on a particular tool — research, images, reasoning or an agent — while ordinary text chat continues.

If an uninterrupted full working day matters, test with your real work before buying a year. Claude Max and ChatGPT Pro buy much more headroom, but neither is a promise of unrestricted use. The published rules are in Anthropic’s usage guide and OpenAI’s current model and limit guide.

Plans and pricing

These are US web prices on 13 August 2026. Tax, currency, app-store billing and regional offers can change the amount. API use and optional usage credits are billed separately.

ChatGPT also has a lower-cost Go plan with regional pricing; Anthropic has no exact plan between Free and Pro. The rows above keep the direct $20, $100 and $200 personal tiers aligned rather than putting unlike plans in the same row.

Fable needs its own pricing note. On Pro and standard Team seats, Fable runs on usage credits at consumption rates rather than using the normal plan allowance. Max and premium Team seats can put up to 50% of their shared weekly limit towards Fable, then continue with credits. The earlier promotion that included Fable in Pro ended on 19 July 2026.

Check the live ChatGPT pricing page, Claude plan guide and Fable plan rules before paying. The $100 and $200 tiers make sense for someone whose work stops when the allowance runs out, not simply for getting nicer answers to occasional questions.

Availability and downtime

Both have outages, and users complain when either is unavailable. That is unsurprising: they are centralised cloud services used by millions of people. Public status pages confirm incidents, but the companies divide their products into components differently, so counting coloured bars is not a fair head-to-head uptime study.

OpenAI published a technical write-up for a 3 February 2026 ChatGPT incident caused by a bad configuration type, and a separate FAQ for its major 10 June 2025 disruption. Claude had a major multi-hour outage on 6–7 April 2026 that affected the web product and API.

There is no sound public evidence that lets me say one consumer service is “frequently down” while the other is not. Check OpenAI Status and Anthropic Status when something fails. If AI access is important to your work, keeping a free account with the other provider is a sensible backup. Shared infrastructure can fail too, so two subscriptions are not the same as guaranteed independence.

Privacy: what happens to your data?

The consumer training default is a meaningful difference. You can opt out in ChatGPT, but you have to do it. Anthropic asks consumers to opt in. Neither policy makes it sensible to paste secrets, client data or personal records into a personal account without checking the settings and the agreement that covers your use.

Read OpenAI’s model-improvement policy and Anthropic’s consumer training policy rather than relying on a screenshot of a settings page. Policies change, and connected apps bring their own data handling into the chain.

Company structure and ethical arguments

OpenAI is controlled by the OpenAI Foundation, with the commercial OpenAI Group organised as a public benefit corporation. Anthropic is also a public benefit corporation and has a Long-Term Benefit Trust whose independent trustees gain authority to elect a growing portion of the board.

Both publish safety frameworks. Anthropic is closely associated with Constitutional AI and a Responsible Scaling Policy for catastrophic risk. OpenAI publishes safety evaluations, system cards and a preparedness framework. These are company policies and governance mechanisms, not independent proof that a product is safe.

Both companies work with the US defence sector. Anthropic’s 2026 dispute with the Department of War centred on two uses it said it would not support: mass domestic surveillance and fully autonomous weapons. OpenAI’s defence agreement states similar red lines around mass domestic surveillance, autonomous weapons and automated high-stakes decisions.

Copyright is another live issue. OpenAI continues to face lawsuits about training data. Anthropic reached a $1.5 billion settlement with authors over pirated book copies in 2026, after a court had separately treated training on lawfully acquired books as fair use. I would not describe either company as free of ethical controversy.

The established difference is in governance, published restrictions and consumer data defaults. A broad claim that one company is “good” and the other is “bad” goes beyond what those facts prove.

Alternatives to ChatGPT and Claude

You do not have to choose either. The realistic alternative depends on what you are trying to replace. The table puts providers along the top so you can scan the trade-offs.

Google Gemini

Gemini is the strongest direct alternative if your working life is in Gmail, Docs, Sheets and Drive. Google AI Pro is listed at $19.99 per month in the US and combines Gemini, Deep Research, large storage and media generation. Google also has serious coding products and NotebookLM. The reason to choose it is integration with Google, not that its chat window looks different.

Microsoft Copilot

Copilot makes most sense when Word, Excel, PowerPoint, Outlook and Teams are already the centre of your work. Researcher can work across organisational files and the web, while image generation and Vision cover consumer jobs. Microsoft can offer models from more than one provider, so choosing Copilot is mainly choosing the Microsoft environment rather than swearing loyalty to one model company.

Perplexity

Perplexity is a search-and-research product first. Paid users can choose current models from OpenAI, Anthropic, Google and others, depending on the plan. It is attractive if you want cited web answers and do not want one model provider. It is less attractive if you want the deepest project memory, coding agent and general computer agent to come from one coherent system.

DeepSeek and Z.ai

DeepSeek and Z.ai offer capable models, low API prices and open-weight releases. DeepSeek also provides OpenAI- and Anthropic-compatible API routes and a coding CLI. Z.ai’s GLM-5.2 is aimed at long-context coding and agent tasks, with a published one-million-token context window.

What you give up is the mature consumer environment around ChatGPT, Claude, Gemini or Microsoft: fewer polished connectors, less general agent infrastructure and a smaller support ecosystem. Their data policies, hosting region and legal jurisdiction may also be different from what your client or employer permits. Check those before sending confidential work, just as you should with a US provider.

So which would I choose?

For a general user paying for one service, I would start with ChatGPT Plus. Native image generation, the wider voice experience, Work and Codex mean it’s a more complete bundle. Its ordinary text limits are also less likely to interrupt a day of short chats.

I would choose Claude Pro instead if my week centred on long documents, Artifacts, terminal coding or giving Cowork a contained folder and a substantial task. I would accept the shared usage allowance and keep another service available for images, as annoying as that is.

For a company already committed to Google Workspace or Microsoft 365, I would price Gemini or Copilot before buying dozens of separate seats. The native connection to the documents people already use can matter more than a small difference between models.

And I would keep a free account with the other main provider. Switching is time-consuming once you have projects, memories, custom instructions, plugins and habits, but it is not a marriage. The healthiest arrangement may be one paid home and one free escape hatch. (Some free, though questionable, relationship advice for you, too!)

This comparison will age probably within a week. Since everything changes all the time, I’ll keep this article up to date using a few tricks I picked up recently.