Claude vs ChatGPT vs Gemini vs Perplexity: which AI model to choose in 2026?

The three launches of early September were not the end of it. Anthropic released Claude Opus 5.5 on 22 September and Sonnet 5.5 on 28 September, OpenAI followed with GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol, and Google announced Gemini 4 Argon on 30 September. If you chose an AI subscription over the summer, it is worth taking another look. We set ChatGPT, Claude, Gemini and Perplexity side by side as they stand on 3 October 2026, with every claim checked with the maker or an independent source. The short answer: there is no winner for everything. Which AI fits depends mostly on your tasks and on the software your team works in.

Want to use AI structurally in your business? ClickForest helps Flemish SMEs in Mechelen, Antwerp and the rest of Flanders through AI for growth, from training to a working application.

Which AI is the best in October 2026?

There is no winner for everything. On the independent Artificial Analysis Intelligence Index, Claude Opus 5.5 stands alone at the top on 3 October 2026 with 58 points, followed by Claude Sonnet 5.5 with 56. Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon share 53 points. On LMArena, where users compare answers blind, Gemini 4 Argon is in first place for now.

For a business, that ranking rarely decides it. What matters more is the software your team works in and what you want to do with the tool. Here is the quick pick:

How to combine the four tools in your work is covered in the superpowers of ChatGPT, Claude, Gemini and Perplexity. A broader overview for marketing teams is in our list of AI tools for marketers.

Which new AI models came out in September 2026?

Plenty. After Claude Fable 5.1 (1 September), Gemini 3.8 Flash (2 September) and GPT-6 Astra (3 September) came Grok 4.7 on 21 September, Claude Opus 5.5 and GPT-6 Sol and Luna on 22 September, Claude Sonnet 5.5 on 28 September and GPT-6.1 Sol on 29 September. Google announced Gemini 4 Argon on 30 September, for now for a small group of users.

The summer was already busy. These are the releases since late June that matter:

Google’s missing Pro model already stood out in early September. Thomas Claburn wrote in The Register:

“Google CEO Sundar Pichai’s reassurances weren’t particularly convincing in light of the company’s failure to release Gemini 3.5 Pro in June as promised.”

— Thomas Claburn, The Register, September 2026

Google itself wrote on 21 July that 3.5 Pro is being tested with partners and will become broadly available “as soon as it’s ready” (Google, July 2026). On 3 October, the Google DeepMind model page still lists 3.5 Pro as “coming soon” (Google DeepMind). In the meantime, Google has already announced Gemini 4 Argon.

What is the difference between ChatGPT, Claude, Gemini and Perplexity?

ChatGPT offers plans from free to Pro and features such as Work, Sites and deep research, Claude is built around agents that finish work on their own, Gemini is built into Google Workspace and also reads video and audio, and Perplexity searches the web and answers with sources. ChatGPT and Claude now also search the web live, so being up to date no longer makes Perplexity unique on its own.

1. ChatGPT: GPT-6 Astra, Work and Sites

OpenAI announced GPT-6 Astra on 3 September 2026 and is rolling it out in phases (OpenAI, September 2026). In the chat it is called GPT-6 Pro, and it is included in ChatGPT Pro (the US$100 and US$200 tiers), Business and Enterprise (help center, October 2026). Plus only gets Astra in ChatGPT Work and Codex. On Enterprise it was off by default at launch (OpenAI). So if you use Plus, you talk to GPT-5.6 Sol in the regular chat. On Free and Go, that is GPT-5.6 Luna.

GPT-6 Sol and GPT-6 Luna followed on 22 September. They are in ChatGPT Work and Codex from Plus, and Free and Go get GPT-6 Luna in the desktop app. OpenAI adds: “These models are not yet available in Chat.” (OpenAI, September 2026). On 29 September, GPT-6.1 Sol started a phased rollout in Work and Codex (release notes).

Independent developer Simon Willison summed up Astra on launch day (simonwillison.net):

“This is clearly OpenAI’s Fable competitor, and appears to score higher than Fable on most of OpenAI’s self-reported benchmarks.”

— Simon Willison, independent open-source developer and creator of Datasette, September 2026

Note the word self-reported: on the independent Artificial Analysis index, Astra and Fable 5.1 are level at 53 points, and Claude Opus 5.5 now sits above both. Greg Kamradt of the ARC Prize Foundation called Astra “the best model we’ve ever tested”, but that quote is on OpenAI’s own launch page.

Beyond the models, ChatGPT competes on features: ChatGPT Work, Sites, deep research, scheduled tasks and a built-in browser (pricing page). Since 29 September, Pro has three usage tiers, including the new Pro 500 at US$500 per month (release notes). The context window depends on your plan, from 27,000 tokens in the free Instant mode to 400,000 in Pro’s reasoning mode. Background on the previous generation is in our ChatGPT 5 guide.

2. Claude: Opus 5.5 as the workhorse, Fable 5.1 for heavy work

Anthropic released Claude Opus 5.5 on 22 September 2026. According to Anthropic, it “performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5” (Anthropic, September 2026). Opus 5.5 is in Pro, Max, Team and Enterprise (Anthropic), and at the same time Anthropic raised the five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans. Sonnet 5.5 followed on 28 September, and according to Anthropic it “runs 30%+ faster” than Sonnet 5 (Anthropic, September 2026). Anyone can use it on claude.ai, including for free (Anthropic).

Simon Willison wrote on launch day:

“The most interesting thing about Sonnet 5.5 is that it’s now the model used for the free tier on claude.ai. OpenAI’s ChatGPT free tier uses Luna 5.6, which means Anthropic currently have a much more capable free offering.”

— Simon Willison, independent open-source developer and creator of Datasette, September 2026

Fable 5.1 came out on 1 September, together with Mythos 5.1. According to Anthropic they are “the same model, but with different levels of safeguards”: Fable 5.1 is generally available, Mythos 5.1 only through programmes for cybersecurity and life sciences, and for now “only available to a set of US organizations” (Anthropic, September 2026).

Fable 5.1 is not in every plan. On Max and on the premium seats of Team, Fable 5.1 is included in your plan, as it is on the premium seats of older (legacy) seat-based Enterprise plans. On Pro you only get it through extra usage credits, and on Free not at all (help center, September 2026). Anthropic’s pages do not say which model is selected by default on Pro or Max; on Enterprise, the admin can set a default model. If you are unsure, Anthropic itself gives this advice: “start with Claude Opus 5.5 for most workloads. Use Claude Fable 5.1 for demanding reasoning and long-horizon agentic work” (Anthropic).

The editors of The Neuron tested Fable 5.1 live and above all noticed a shift in how much you hand over to such a model:

“The surprise was how quickly the question changed from “Can it do this?” to “How much judgment should we let it exercise without asking us first?””

— The Neuron editorial team, September 2026

Claude is strongly focused on work an agent finishes on its own. Claude Cowork completes tasks in your files, Claude in Chrome operates your browser and Claude Code writes software (see vibe coding versus Claude Code). How the desktop app bundles those modes is covered in Claude Desktop explained. The context window is large: on paid plans, Fable 5.1, Opus 5.5, Sonnet 5.5 and their predecessors Opus 5 and Sonnet 5 have 1 million tokens in the chat (help center). Anthropic publishes no figure for the free version. And Claude reads text and images, not video or audio.

3. Gemini: 3.8 Flash and deep inside Google Workspace

Google calls Gemini 3.8 Flash, released on 2 September 2026, “our most intelligent workhorse model” (Google, September 2026). If you have Google AI Pro or Ultra, you get it in the Gemini app, in AI Mode in Google Search and in Gemini in Google Sheets. At launch, Artificial Analysis called it “the cheapest model at its level of intelligence”, according to The Register.

The Pro model in the app is still Gemini 3.1 Pro, which carries the Preview label (Google DeepMind). On the independent index it scores lower than 3.8 Flash: 30 against 41 points. Free users get 3.6 Flash and limited access to 3.1 Pro.

Google’s new top model already exists, but you cannot use it yet. Gemini 4 Argon, announced on 30 September, is going for now to “a set of trusted cyber defenders through our Fairwind Program” (Google, September 2026). According to Google, the broader rollout comes later, “starting with paid API customers and Google AI Ultra subscribers”. Artificial Analysis itself: “Gemini 4 Argon is currently being rolled out to selected users and is not publicly available” (Artificial Analysis, September 2026).

Its biggest strength is integration. Gemini is part of the Google Workspace plans, and from Starter up you get the AI assistant in Gmail and the Gemini app (Workspace). Gemini also reads text, images, video, audio and PDF, with a context window of 1 million tokens in Google AI Pro. For going through recordings, that is a real difference from Claude.

4. Perplexity: search with sources, on other companies’ models

Perplexity does not build a top model of its own. It combines its own model, Sonar 2, with models from others. In Pro you choose from, among others, GPT-5.6 Terra, Claude Sonnet 5 and a Gemini Flash model; Max adds GPT-5.6 Sol and Claude Opus 5 (Perplexity). That list dates from 4 September and does not yet mention Anthropic’s new 5.5 models. Perplexity itself says the model selector in your account is the only reliable reference. If you search for free, you get the automatic “Best” mode and a limited number of Pro searches.

The Comet browser has been free since 2 October 2025 (Perplexity), and since March 2026, according to Perplexity, Pro users also have access to Perplexity Computer, Perplexity’s agent system (changelog). How Comet compares to Anthropic’s browser extension is covered in Claude in Chrome explained. For the day-to-day approach per tool, see our practical guide to ChatGPT, Claude and Perplexity.

5. Grok: the outsider with 500,000 tokens of context

xAI released Grok 4.5 on 8 July 2026, Grok 4.6 on 12 August and Grok 4.7 on 21 September, all three with a context window of 500,000 tokens (xAI). Via the API, Grok 4.5 and Grok 4.7 cost US$2 per million input tokens and US$6 per million output tokens, as long as your prompt stays under 200,000 tokens (xAI, release notes). On the independent index, Grok 4.7 scores 46 points, below the top models from Anthropic, OpenAI and Google.

What do the benchmarks say about ChatGPT, Claude and Gemini?

On the Artificial Analysis Intelligence Index v4.3.2, Claude Opus 5.5 stands alone at the top on 3 October 2026 with 58 points, followed by Sonnet 5.5 with 56. Fable 5.1, GPT-6 Astra and Gemini 4 Argon share 53 points, and GPT-6.1 Sol scores 52. On LMArena, where users compare answers blind, Gemini 4 Argon is in first place for now (Artificial Analysis, LMArena).

Each model shows the best-scoring variant listed on each platform, with the LMArena standings of 2 October 2026. For Opus 5.5 and Sonnet 5.5, the Artificial Analysis score is for the “with fallback” variant: for flagged requests, an older model answers. Artificial Analysis works with points out of 100, not percentages. The same number of points does not cost the same either. In early September, Artificial Analysis wrote about this:

“Similar Intelligence Index scores can come at very different costs. GPT-6 Astra (max) and Claude Fable 5.1 (max with fallback) both score 53, but their average cost per Intelligence Index task is $3.26 and $7.63 respectively - 57% lower for Astra.”

— Artificial Analysis, in Intelligence Index v4.3 (7 September 2026)

Read those figures with some caution. The index changes version regularly (v4.3 in early September, v4.3.2 now), so never compare scores from different versions. The makers each publish their own benchmarks with their own test setups: OpenAI reports 99.9% on ARC-AGI-3 with its own harness, while according to Simon Willison it was 62.7% with the standard setup. And on LMArena the newest models still have few votes: 4,932 for Gemini 4 Argon, 4,552 for Opus 5.5. Their position may still shift.

At the launch of Opus 5.5, Anthropic itself added a caveat:

“That said, at these levels of capability we’ve found that benchmark margins have become a less reliable guide to real-world differences.”

— Anthropic, September 2026

A benchmark measures how a language model scores on a fixed set of tasks; how it handles your own tasks is something you are best off testing yourself, with good prompts.

Which tool suits your team depends on your own work. ClickForest helps Flemish SMEs in Mechelen and Antwerp with AI training for the whole team.

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

For one user in Belgium you pay €23 per month for ChatGPT Plus, €18 for Claude Pro (monthly, excluding VAT, or €15 when billed annually), €21.99 for Google AI Pro and US$17 for Perplexity Pro when billed annually. It can be cheaper too: Google AI Plus costs €4.99 and ChatGPT Go €8. All four have a free version.

These are monthly prices as shown from Belgium on 3 October 2026 (ChatGPT, Claude, Gemini, Workspace, Perplexity, Perplexity Enterprise). Anthropic states that VAT is not included, OpenAI and Google do not say, and Perplexity only shows dollars. For Workspace these are regular prices; temporary discounts vary.

What you get for that money differs too:

• GPT-6 Astra is in the chat (as GPT-6 Pro) from ChatGPT Pro and in Business; Plus only gets Astra, GPT-6 Sol and GPT-6 Luna in ChatGPT Work and Codex.
• Claude Opus 5.5 is in Pro, Max, Team and Enterprise, not in Free. Fable 5.1 is included on Max and on the premium seats of Team and of older (legacy) seat-based Enterprise plans; on Pro you only use it through extra credits.
• You get Gemini 3.8 Flash from Google AI Pro, and AI Ultra adds Deep Think.
• In Perplexity you only choose GPT-5.6 Sol and Claude Opus 5 from Max.

A comparison of the features of ChatGPT, Claude and Perplexity per plan is in our AI tools comparison.

Which AI works best with Google Workspace or Microsoft 365?

In Google Workspace that is Gemini: from the Starter plan at €6.80 per user you get Gemini in Gmail and the Gemini app. In Microsoft 365 you work with Copilot, where since 22 September Microsoft has offered Claude Opus 5.5 and GPT-6 Sol in Word, Excel, PowerPoint, Chat, Cowork and Copilot Studio. Since 30 September, Sonnet 5.5 and GPT-6.1 Sol have been rolling out as well.

That Microsoft rollout happens in phases (Microsoft, 22 September 2026). Sonnet 5.5 and GPT-6.1 Sol came first to Copilot Cowork and Copilot Studio, with usage-based billing; according to Microsoft, Word, Excel, PowerPoint and Chat follow “in the coming week” (Microsoft, 30 September 2026). When Claude Opus 5 arrived in July, Microsoft already wrote: “Availability may vary by region and tenant configuration” (Microsoft, July 2026). So check with your IT partner what is already switched on in your environment.

For Workspace customers, privacy counts too. Google writes that your content “is not human reviewed or otherwise used for Generative AI model training outside your domain without permission” (Google Workspace). If your team uses neither suite, a standalone subscription to ChatGPT or Claude is often simpler than adding a full office suite.

Do ChatGPT, Claude, Gemini and Perplexity use your data for training?

On the consumer plans they do by default, unless you switch it off: that applies to ChatGPT Free, Go, Plus and Pro, to Claude Free, Pro and Max and to Perplexity Free, Pro and Max. Google uses chats from personal accounts to improve its models. On the business plans, none of the four trains on your business data by default.

That is how it is stated on the pricing pages of OpenAI and Anthropic, in the Gemini app privacy notice, at Google Workspace and on Perplexity’s data page.

The practical rule: do not put personal data or confidential business information into a consumer plan, and take a business plan as soon as your team works with customer data. For customers’ personal data, the GDPR applies as well.

Claude, ChatGPT and Gemini are officially available in Belgium (Anthropic, Google), Gemini also in Dutch and French, and OpenAI confirms GPT-5.6 for the whole EEA (OpenAI). From Belgium, Perplexity simply shows its plans, in dollars. Claude Mythos 5.1 is the exception: for now only for a limited group of US organisations. Under the EU AI Act, Anthropic adds a watermark to output from models released after 2 August 2026, Opus 5.5 included; a detection API is still in private preview (Anthropic, Anthropic).

And apart from privacy: no model is free of hallucinations. Always have figures, quotes and legal content checked by a person.

How do Belgian businesses use ChatGPT, Claude, Gemini and Perplexity together?

By using each tool where it is strongest: Perplexity for external research with sources, Gemini for your own files in Workspace, and Claude or ChatGPT for writing, analysing and building. According to Stanford’s AI Index Report 2026, 88% of organisations now use AI (Stanford HAI).

This is how we combine the tools at ClickForest in marketing work. How AI fits into your marketing more broadly is covered in the AI marketing guide and in our piece on AI agents in marketing.

1. Market research: Perplexity and Gemini

Ask Perplexity for an overview of your five biggest competitors in Belgium, with their offer and prices, and you get a table with sources. Then put that in Gemini next to your own sales figures in Google Sheets and ask where you are leaving margin on the table. That way you combine current market information with your own data, without copying that data into an extra tool.

2. Content and campaigns: Claude or ChatGPT, with Gemini

Let Gemini analyse the visuals of your best-performing ads, and give that analysis to Claude or ChatGPT to write ten variants of your ad copy. Test the variants, keep what works and turn the rest into content for social media.

3. Shopify webshop: Claude and Gemini

Export your orders as a CSV and let Claude analyse which products are often bought together. Then let Claude write the code to offer those products as a bundle, and Gemini the product copy in your brand’s tone. Need help with the technical side of your Shopify webshop or with conversion optimisation? ClickForest can help with that.

4. SEO and GEO: Perplexity and Gemini

Use Perplexity to see which questions people ask about your niche and which sources AI answers cite. Then let Gemini or Claude build content based on your own expertise that answers those questions. How you yourself end up in the answers of ChatGPT, Claude, Gemini and Perplexity is the domain of GEO.

Conclusion: which AI should you choose?

After the September releases, Claude Opus 5.5 stands alone at the top of the independent index, with Sonnet 5.5 right behind it. Via the API, Gemini 3.8 Flash still costs a fraction of GPT-6 Astra or Fable 5.1, and for now Gemini 4 Argon only goes to a small group of users.

Our advice: choose first on your software and your tasks, and only then on the ranking. If you work in Google Workspace, start with Gemini. If you work in Microsoft 365, check which models Copilot already offers you. If you do a lot of research, add Perplexity. If you want AI to finish work on its own, Claude with Cowork and Claude in Chrome is a good place to start. And only put personal data in a business plan.

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