ChatGPT vs Claude vs Perplexity: which AI tool wins in 2026?

ChatGPT, Claude and Perplexity are three AI tools with their own strengths, and they change faster than any comparison can keep up with. Canvas has not been in ChatGPT’s standard models since late May, GPT-6 Astra arrived on 3 September, Claude in Chrome has been generally available since 26 August and Perplexity Computer now also works on Pro. Here we put the three tools side by side again, as of 12 September 2026, with every claim checked with the maker or an independent source.

What is the difference between ChatGPT, Claude and Perplexity?

ChatGPT bundles chat, image generation, voice, deep research and the ChatGPT Work agent in one app. Claude is built around agents that finish work on their own, such as Cowork and Claude in Chrome, and has no built-in image generator. Perplexity is a search engine that answers with sources, with its own model, Sonar 2, and models from OpenAI, Anthropic and Google.

ChatGPT from OpenAI had more than 900 million weekly users and more than 50 million paying subscribers at the end of March 2026, according to OpenAI (OpenAI, March 2026). On Free and Go you work with GPT-5.6 Luna, from Plus with GPT-5.6 Sol, and since September GPT-6 Astra has been included in Pro, Business and Enterprise (OpenAI).

Claude from Anthropic focuses on work that an agent completes: Claude Cowork finishes tasks in your files, Claude in Chrome operates your browser and Claude Code writes software. Since 1 September the top model has been Claude Fable 5.1, alongside Opus 5, Sonnet 5 and Haiku 4.5.

Perplexity gives an answer with citations for every search. It combines its own search model, Sonar 2, with models from OpenAI, Anthropic and Google, and the Comet browser has been free since October 2025. In SEO work and GEO it quickly shows you which sites Perplexity cites as a source.

That such tools can save time became clear early on. In an experiment by Shakked Noy and Whitney Zhang of MIT (Science, 2023), the time spent on writing tasks fell by 40% and quality rose by 18%, among 453 college-educated professionals (MIT News, July 2023). Noy added a caveat himself:

“The experiment demonstrates that it does bring significant speed benefits, even if those speed benefits are lesser in the real world because you need to spend time fact-checking and writing the prompts.”

— Shakked Noy, PhD student in economics at MIT, July 2023

ClickForest guides Flemish SMEs in Mechelen and Antwerp in choosing and using such tools, through AI for growth.

Which features do ChatGPT, Claude and Perplexity have?

All three search the web live with sources, remember context across conversations and work with projects. The big differences lie elsewhere: ChatGPT and Perplexity generate images, Claude has no built-in image generator; Perplexity also makes short videos; and each has its own approach to agents, from ChatGPT Work to Claude Cowork and Perplexity Computer.

Canvas, ChatGPT’s separate writing and coding space, has not been in the standard models since 28 May 2026. OpenAI wrote at the time: “Writing and coding functionality is now supported directly in chat responses through writing blocks and code blocks”. Paying users could still use canvas for a limited time through older models then (OpenAI). And Claude does search the web: web search has been on all plans since May 2025 (Anthropic), and according to Anthropic: “Every response includes citations” (Claude Help Center).

Ethan Mollick, professor at Wharton, sums up the difference in images briefly: “Both Google and ChatGPT have really great image generators built in; Claude has none” (One Useful Thing, July 2026).

Which AI models are in ChatGPT, Claude and Perplexity?

ChatGPT runs on GPT-5.6 Luna on Free and Go and on GPT-5.6 Sol from Plus; since September GPT-6 Astra has been in Pro, Business and Enterprise. Claude has Fable 5.1 as its top model, alongside Opus 5, Sonnet 5 and Haiku 4.5. Perplexity combines its own model, Sonar 2, with models from OpenAI, Anthropic, Google and others.

According to Anthropic, Claude Fable 5.1 and Mythos 5.1 are the same model with different safeguards; for now Mythos is only available to a limited group of US organisations (Anthropic). Opus 5, Sonnet 5 and Fable 5.1 decide for themselves how much to think (adaptive thinking). If you are unsure, Anthropic gives this advice: “start with Claude Opus 5 for most workloads. Use Claude Fable 5.1 for demanding reasoning and long-horizon agentic work” (Anthropic).

The line-up in Perplexity’s model picker changes regularly; on 12 September the Pro page already showed Gemini 3.8 Flash (Perplexity). On the independent Artificial Analysis index, Claude Fable 5.1 and GPT-6 Astra share first place with 53 points (as of 7 September 2026). How the models score beyond that, and how Gemini and Grok compare, is covered in our comparison of Claude, ChatGPT, Gemini and Perplexity.

When do you use ChatGPT, Claude or Perplexity?

Use ChatGPT if you want one app for writing, images, voice and research, Claude if you want an agent to finish work for you or you build with code, and Perplexity if you need answers with verifiable sources. Ethan Mollick, professor at Wharton, sees only two choices for people who want to get the most out of AI:

“There are only two choices for most people who want to get the most out of AI right now: ChatGPT or Claude”

— Ethan Mollick, professor at Wharton, July 2026

ChatGPT for content, images and longer tasks

Content and images in one app. ChatGPT writes, generates images with Images 2.0 on all plans and works with voice. If you create a lot of content with AI, you have text and images in one place.

Longer tasks with ChatGPT Work. According to OpenAI, Work is “an agent designed for longer, multi-step work and finished deliverables” (OpenAI). It replaces the former agent mode (OpenAI) and is still rolling out gradually. Simon Willison does find its behaviour confusing:

“I think the difference between ChatGPT Work on a mobile device and ChatGPT Work inside the desktop app (where it’s effectively a less intimidating skin on top of Codex) is spectacularly unintuitive.”

— Simon Willison, independent developer and blogger, July 2026

Code with Codex. For programming, ChatGPT has Codex. For now, Plus gets GPT-6 Astra only in Work and Codex, not in the regular chat.

Claude for agents, code and long documents

Work an agent finishes on its own. Cowork completes tasks in your files and Claude in Chrome operates your browser. That suits recurring work in marketing automation and AI agents in marketing.

Code and custom work. With Claude Code you build software. ClickForest uses it every day to build websites, CRM and custom digital tools.

Long documents, with one caveat. Claude reads PDFs of up to 1,000 pages per file and up to 20 files per chat (Claude Help Center), but in the apps the context window is 200,000 tokens (500,000 on Enterprise); the million tokens only applies via the API.

Perplexity for research with sources

Research with sources. According to Perplexity, Research, the successor to Deep Research, completes most tasks in under three minutes (Perplexity).

“Perplexity positions itself as a real-time answer engine to provide concise, cited responses. This emphasis on provenance can appeal to users who are seeking verifiable information, such as researchers and executives.”

— Darren Kimura, President and CEO of AI Squared, June 2025

Market data, images and video. Perplexity has a separate Finance environment with company and market data. On Pro and Max it also generates images, and short videos of up to eight seconds with Veo 3.1 (Perplexity).

What are the pros and cons of ChatGPT, Claude and Perplexity?

ChatGPT has writing, images, voice and an agent in one app, although the free version only allows three uploads per day. Claude focuses on agents and code with Cowork, Claude in Chrome and Claude Code; it does not have a built-in image generator. Perplexity gives sources with every answer. Legally, though, it is under fire from publishers.

ChatGPT

Pros:

• Writing, images (Images 2.0), voice and deep research in one app
• ChatGPT Work for longer tasks with a finished result
• Unlimited text chat in the free version, with GPT-5.6 Luna
• More than 900 million weekly users according to OpenAI, so plenty of guides and examples

Cons:

• Canvas has not been in the standard models since May 2026
• The free version only allows three uploads per day (OpenAI)
• Plus gets GPT-6 Astra only in Work and Codex; in the regular chat it is in Pro (from €103 per month), Business and Enterprise
• Full MCP integrations only on Business and Enterprise

Claude

Pros:

• Agents that finish work on their own: Cowork and Claude in Chrome
• Web search with sources on all plans
• Integrations via MCP with remote connectors, also on the free plan (one connector) (Claude Help Center)
• Voice mode on all plans, still in beta

Cons:

• No built-in image generator (Claude Help Center)
• 200,000 tokens of context in the apps, the million only via the API
• On Pro, Fable 5.1 is only available through extra credits

Perplexity

Pros:

• Citations with every answer
• In Pro a choice of models from OpenAI, Anthropic and Google
• Free Comet browser, and Perplexity Computer on Pro

Cons:

• Uploads of up to 40 MB per file (Perplexity)
• Prices only shown in dollars and for annual billing
• Lawsuits from publishers, including The New York Times since December 2025 (TechCrunch)
• Claude Opus 5 and GPT-5.6 Sol only from Max

What do ChatGPT Plus, Claude Pro and Perplexity Pro cost?

In Belgium, ChatGPT Plus costs €23 per month. Claude Pro costs €18 per month, or €15 per month when billed annually, excluding VAT. Perplexity Pro costs US$17 per month when billed annually. All three have a free version, and for heavier use there are ChatGPT Pro from €103, Claude Max from €90 and Perplexity Max at US$167 when billed annually, all per month.

These are the prices as shown from Belgium on 12 September 2026 (ChatGPT, Claude, Perplexity). Anthropic states that VAT is not included, OpenAI does not say, and Perplexity only shows dollars. In the US, ChatGPT Plus costs US$20 per month (OpenAI).

ChatGPT Pro has two levels, of US$100 and US$200 per month. According to OpenAI, since 10 September 2026 you temporarily cannot sign up for or upgrade to the US$200 plan (OpenAI). Mollick also warns about the standard plans: “the $20 tiers include real but limited agent usage, and agents burn through those limits quickly”.

Which AI tool do you choose for which task?

Choose per task: Perplexity for research and market data with sources, ChatGPT for content with images and voice, and Claude for code, analysis and work an agent finishes on its own. Start with one paid plan and only add a second when a task really calls for it.

An example from marketing work: start the research in Perplexity, let Claude analyse the findings alongside your own figures, and create the visuals and variants in ChatGPT. How to use each of the three in practice is covered in our practical guide to ChatGPT, Claude and Perplexity. Whichever tool you choose, the result depends heavily on your prompt: read write better prompts or the explanation of the RACE model.

Want to introduce this approach in your team? ClickForest gives Flemish SMEs in Mechelen and Antwerp AI training for the whole team.

What changed in 2026 at ChatGPT, Claude and Perplexity?

OpenAI removed older models from ChatGPT, took canvas out of the standard models, replaced agent mode with ChatGPT Work and released GPT-6 Astra. Anthropic launched the Claude 5 generation, Fable 5.1, Cowork on web and mobile and Claude in Chrome for all paid plans. Perplexity made Computer available on Pro and replaced the Sonar API with the Agent API.

OpenAI

• February to August: GPT-4o, GPT-4.1, GPT-5, GPT-5.1, GPT-4.5 and o3 disappear from ChatGPT (OpenAI, release notes)
• 28 May: canvas disappears from the standard models, writing and code now happen in the chat itself
• 9 July: GPT-5.6 (Sol, Terra, Luna) and the move to the Plugin directory
• 3 September: GPT-6 Astra
• 10 September: new sign-ups and upgrades to the US$200 ChatGPT Pro plan temporarily paused

Anthropic

• 30 June and 24 July: Claude Sonnet 5 and Claude Opus 5
• July: Cowork also works on web and mobile
• 26 August: Claude in Chrome generally available on all paid plans
• 1 September: Claude Fable 5.1, together with Mythos 5.1 for a limited group of US organisations

Perplexity

• February: the Agent API generally available; in July it replaces the Sonar API (Perplexity)
• 12 March: Perplexity Computer for Pro users (Perplexity)
• 4 August: a US court of appeals overturns the preliminary injunction Amazon had obtained against the Comet browser (Ninth Circuit)

Usage is growing too. According to Stanford’s AI Index Report 2026, 88% of organisations now use AI (Stanford HAI). Among US companies, according to Menlo Ventures, 40% of spending on language models went to Anthropic at the end of 2025, 27% to OpenAI and 21% to Google, based on API usage (Menlo Ventures, December 2025). That figure is about business API spending in the US; it does not measure which chat app a Belgian SME uses.

Which AI browsers are there in 2026?

Perplexity Comet is a separate AI browser, free since October 2025. ChatGPT Atlas is stopping: OpenAI announced this on 9 July 2026 and scheduled the end for 9 August 2026. Claude in Chrome is an extension in your existing browser, generally available on all paid Claude plans since 26 August 2026. And Arc is no longer actively developed.

Perplexity Comet (launched in July 2025, free since 2 October 2025):

• A separate browser with Perplexity built in, for Mac, Windows, iOS and Android (Perplexity)
• At launch first for Max subscribers and through a waitlist (TechCrunch), free for everyone since 2 October 2025 (Perplexity)

ChatGPT Atlas (released on the Mac in October 2025):

• On 9 July 2026 OpenAI announced that Atlas is stopping (9to5Mac); OpenAI scheduled the end for 9 August 2026
• The browser features move to the ChatGPT desktop app and, where available, a ChatGPT extension or sidebar for Chrome (OpenAI)

Claude in Chrome (generally available since 26 August 2026):

• An extension that reads pages, clicks, types and fills in forms with your existing logins
• Only in Google Chrome on a computer; permissions and security are covered in Claude in Chrome explained

Arc (The Browser Company):

• According to Engadget, The Browser Company stopped active development of Arc in May 2025, apart from security updates
• Atlassian announced the acquisition of The Browser Company in September 2025, for US$610 million (CNBC)

For anyone working on SEO and GEO, it matters that a visitor can also view your site through such an agent. How webshops prepare for that is covered in our piece on agentic commerce; visibility in AI answers itself is the domain of GEO.

Conclusion: the right tool for the right job

There is no absolute winner between ChatGPT, Claude and Perplexity. ChatGPT bundles writing, images, voice and an agent, Claude focuses on agents that finish work on their own, and Perplexity remains the tool for research with sources.

Our recommendation: choose one main tool based on your daily work, ChatGPT or Claude, and add Perplexity if you do a lot of research. Start with the free versions, test them for a week in your own workflow and only pay for what really saves time. Only put customer or company data into a business plan.

Want to know how Gemini and Grok compare? That is covered in our comparison of Claude, ChatGPT, Gemini and Perplexity. An overview of niche tools per task can be found in the best AI tool per task.

ClickForest helps Flemish SMEs and scale-ups in Mechelen, Antwerp and beyond make that choice and build AI into their work, with AI training, AI agents and building with AI. Want to talk about your situation? Get in touch.