‘Which AI models should I use? Should I get a subscription to one, or more?’ The answer is: model surfing. Rather than committing to a premium subscription for a single large player, we look at the strategic use of multiple free accounts across different platforms, maximizing free token allowances and compare functionalities before making any financial commitment. AI subscription services for non-technical professionals can add up. We compare pricing tiers and features across the frontier models and note that subscription prices are trending downward due to competition, particularly from free upstarts like DeepSeek.

AI Leadership – Fiona Passantino, 27 FEB 2025

Pay to Play

You’ve been using AI for a few weeks or months now. Maybe you’ve had some training, watched a few YouTube videos, or taken advice from a podcast. You’ve quietly woven it into your working routines, with or without your manager’s blessing, or the awareness of your friends and colleagues. Congratulations; you’re now officially an AI user.

You’re building your stack, experimenting with various platforms. Perhaps starting with ChatGPT or sticking with your company’s Copilot-only environment. You’re a typical non-technical professional, using AI via a browser or an app. You’re not writing code, nor fitting an API* nor developing an internal SLM**. And you’re noticing that premium memberships to AI services are starting to add up.

The Lineup

One to the questions I get asked most frequently is: “Should I get a premium membership for ChatGPT? For Copilot? Gemini? Which one, and at which level?” There’s no good answer to that question, and the playing field is changing all the time.

Most frontier models offer some form of free access that allows for a limited number of tokens to be processed and then offers a tiered system for heavier use.

Here’s a snapshot of the state of the AI jungle as of right now.

ChatGPT

GPT Plus is $20/month. What does that buy you? Extended limits on messaging, file uploads, data analysis and the number of images you’re able to generate[i]. You can chat by app, by web and using advanced voice mode with video and screensharing, even if European users do not have access to the best features (why do Europeans get the table scraps, later than the American users? Could be a strategic move to lock this huge potential competitor out of the market or just a Tech Bro Tantrum to protest European AI regulation).

In addition, ‘Plus’ gives you access to multiple reasoning models: o3‑mini, o3‑mini‑high, and o1. You also get projects, tasks and custom GPTs (or ‘agents’). There is limited access to their Sora video generation.

GPT ‘Pro’ is their top user tier. This offers unlimited access to all reasoning models, including GPT‑4o, Deep Research, and the o1 pro mode (more compute for the best answers to the hardest questions). You have unlimited access to advanced voice mode and higher limits for video, screensharing, Sora video, and the research preview of Operator (but, only for those living in the US!). This costs a whopping $200/month.

Claude

The ‘kinder, gentler’ and perhaps safer AI frontier model, Claude from Anthropic offers a ‘Pro’ version with more tokens for their ‘Projects’ platform, access to their 3.5 Sonnet and Claude 3 Opus, as well as a glimpse into new features before the others get hold of them. ‘Pro’ is $18/month[ii].

One tier above that is ‘Team’ which is $25/month and offers more usage than Pro, central billing and some advanced administration features. You also get early access to new features before they are launched in the wider world.

Gemini

Like the others, the standard Gemini AI is free to use, straight from the browser. Open an account and get extra credit when you use your GMail and Chrome browser. ‘Advanced’ is $19.99/month. ‘Business’ starts at $20/month, per person, for a 1-year commitment. ‘Enterprise’ starts from $30/month, per user, for a 1-year commitment.

What’s great about Gemini is the ability to use these tools straight in your Google productivity apps; your Gmail, Google Docs, Google Meet and more. If your organization uses these tools, it makes a Gemini subscription that much more interesting.

Microsoft Copilot

The beauty of Microsoft is not in its performance, but the fact that its productivity apps are used by most of the computers of the world (see Copilot for more). ‘Pro’ gives you preferred access to the latest AI models and more tokens at $19.00 per user per month. Copilot AI agents can be added to just about all your 365 apps, such as Word, Excel, PowerPoint, OneNote and Outlook[iii]. There is extensive use of ‘Designer’, the image generator, and you can interact with the models at any time of day or not, not just the ‘off-peak’ times, and do all of this with the Copilot Voice tool.

Mistral

Mistral is Europe’s own AI, promising its users on the continent a good experience done quickly. A Pro account is $14.99/month, which offers unlimited web browsing, an unlimited number of messages per day, extended access to news, file uploads, data analysis, and image generations It also offers ‘Flash answers’[iv].

What are ‘Flash answers’? Not any specific type or service, platform or interface, it’s just Mistral’s friendly way of telling you that their bot est trés vite and delivers a good, low-latency answer in pas de temps). Mistral ‘Team’ is $24.99 and offers even more tokens, live AFP news, file uploads, advanced data analysis, image generation, and even faster Flash answers.

Grok

Elon Musk’s ego baby is a cheap, good model, build to stick to the Man (Sam Altman from OpenAI), so if you’re OK with Elon’s politics and other strange behavior, take advantage of the War of the Bros by opening a Grok account. Basic Grok is $3 per month, Premium is $8 per month, and Premium+ is $40 per month. While the price is still within the reservation when compared with the others, it represents an 81% increase on the monthly plan for the top tiers from a year ago[v].

DeepSeek

Free. This newcomer disrupter Chinese AI model launched with the purpose of creating a stir and unseating the Frontier giants from their high perches. ‘Free’, however, is not without cost. It operates under strict Chinese data laws covering cybersecurity and data security, requiring data storage within China and potential access to your information by the Chinese government.

This raises privacy risks if you are doing espionage abroad and might be problematic for confidential business or personal data, which may not comply with GDPR or CCPA. Security concerns might make you feel squirmy when it comes to state oversight, potential surveillance and data interception, making sensitive information vulnerable. You will likely get a slanted view of the news. Just try typing in anything about Tiananmen Square and see what happens. This, and other Chinese models, actively filter or censor content which might affect your research or decision-making goals.

But all AI models are weighted in their own ways, driven by the culture, laws and interests of the countries in which they operate. If you’re OK with sending your non-sensitive emails over to Shanghai and feel safety when running with the giant herd of Wildebeest that represents user data, go ahead and enjoy the benefits of fast, good and free.

Here’s the bottom line.

Model	Tier 1	Tier 2	Tier 3
GPT (OpenAI)	$20 (Pro)	$200 (Plus)	You can pay more??
Claude (Anthropic)	$18 (Pro)	$25 (Team)	Call and ask (Enterprise)
Gemini (Google)	$19.99 (Advanced)	$20 (Business)	$30 (Enterprise)
Copilot (Microsoft)	$20 (Pro)	$25 (Pro Business)	Call and ask (Enterprise)
Le Chat (Mistral)	$14,99 (Pro)	$24.99 (Team)	Call and ask (Enterprise)
Grok (XAI)	$3 (Basic)	$8 (Premium)	$40 (Premium+)
DeepSeek	‘Free’	Still free…

The Solution

The answer to the earlier question – what model to commit to – is… none of the above. But engage is my new term: “model surfing”. This is the practice of having a range of free accounts, navigating through various AI platforms, using each account’s daily free token allocation until they dry up. And then jumping over to the next one when they do.

Aside from saving money, using multiple AI models exposes you to their unique strengths. This allows you to assess and compare functionalities before you’re tempted to sign a contract with one of them.

What’s a Token?

To be a shrewd model surfer, it’s good to understand what a ‘token’ is in the world of AI. A ‘token’ is the smallest unit of data that is processed by AI language models. They represent the Lego bricks of text, such as words or characters, or the building blocks of images – the raw pixel – that the model interprets to generate responses.

Each interaction with an AI consumes a certain number of tokens, making up both the input (your prompt) and the output (AI’s response). For instance, OpenAI’s GPT-4 model offers a context window of up to 8,192 tokens for free users, while premium subscribers get substantially more[vi]. These are refreshed every day.

The Future

According to McKinsey, 92% percent of companies plan to increase their AI investments over the next three years[vii].But what are they investing in? Off the shelf AI or proprietary models?For individual users and small businesses, these costs can be prohibitive.

Prices for AI models are coming down. The DeepSeek effect is already working its magic across the Frontier; its models are 40 times lower than the GPT-suite at OpenAI[viii]. This caused an immediate effect: Altman’s strategic release of the GPT o3 model a free tier just three weeks later.

So, get out there and surf. The waters may be choppy and constantly changing, which makes it a great time to get your feet wet without having to commit.

---

* API, or ‘Application Programming Interface’, is a set of rules and instructions that allow different software applications to communicate with each other, so that one system (for instance, a client request) can access and use the functionality or data of another (for instance, the business provider or platform serving the client). Read more about APIs.

** SLM, or ‘Small Language Model’ is a type of generative AI application that operates on a much smaller scale than the giant frontier models like ChatGPT, Meta or Copilot. It’s small because it’s not trained on the full extent of our Human knowledge over millennia, but rather on a highly specialized, limited dataset that’s specifically curated for a particular area of expertise (for instance, legal, biology, company-specific information, teachers’ tools etc.).

Need help with AI Integration?

Reach out to me for advice – I have a few nice tricks up my sleeve to help guide you on your way, as well as a few “insiders’ links” I can share to get you that free trial version you need to get started.

[i] OpenAI (2025) “Pricing Tiers”, OpenAI website. https://openai.com/chatgpt/pricing/

[ii] Anthropic (2025) “Pricing” Anthropic website. https://www.anthropic.com/pricing

[iii] Microsoft (2025) “Copilot Pro” Microsoft website. https://www.microsoft.com/en-us/store/b/copilotpro

[iv] Mistral (2025) “Everything you need at the frontier” Mistral website. https://mistral.ai/products/la-plateforme#pricing

[v] Marshall (2025) “Elon Musk Hikes X Premium+ Plan Following Grok 3” Tech.co. https://tech.co/news/x-premium-price-increase-grok-3

[vi] SpicyChat.ai (2024) “Token Limits and AI Models” Spicy Chat webiste https://docs.spicychat.ai/product-guides/creating-chatbots/token-limits-and-ai-models?utm_source=chatgpt.com

[vii] Mayer, Yee, Chui , Roberts (2025) “Superagency in the workplace: Empowering people to unlock AI’s full potential” McKinsey 2025 Report, McKinsey Digital. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work

[viii] Soni, Kachwala (2025) “DeepSeek’s low-cost AI spotlights billions spent by US tech” Reuters. https://www.reuters.com/technology/artificial-intelligence/big-tech-faces-heat-chinas-deepseek-sows-doubts-billion-dollar-spending-2025-01-27/