My AI Stack: Which Model to Use When (And Why)

AI with ALLIE

The professional’s guide to quick AI bites for your personal life, work life, and beyond.

Picking the best AI model for the job becomes easier with time and practice. But because I know many of you are overworked, underappreciated, and looking for learning curve accelerations, here’s a straight whiskey shot of AI usage tips.

Let's jump into it 👇

When to Use Each Top AI Model Right Now 💡

Ever wonder which AI brain to tap for different tasks? I’ve now taught AI on all 7 continents, and one of the questions I get most often is surprisingly simple:

"Which AI should I actually be using?"

The landscape is shifting faster than TikTok trends, with new models and features dropping almost daily. I wanted to wait to share this list until after GPT-4.5 was released to Plus users—and since that happened yesterday (huzzah!), I can now proudly showcase, after thousands of hours testing and speaking with each of the major AI players, my own opinions about which AI to use when and where they all fall flat.

🚨 Before the pitchforks come, this is only my opinion. The answer is also an unfortunate combination of traits: it’s subjective, it’s constantly changing, and it’s a longer answer than folks want to hear. Your AI mileage may vary.

Let me break down my personal AI stack so you can stop wasting time with the wrong tool for the job.

The Big Three: Who's Really Worth Your Time?

Let's get something out of the way immediately: you can absolutely get by with just ONE of the top three assistants (ChatGPT, Gemini, or Claude). And if you're a casual user who doesn’t need the latest model or feature immediately on tap, the free versions will likely cover your bases. You don’t have to listen to your annoying cousin telling you to trial 500 AI tools and measure the pixel-by-pixel performance of each. If you just want to create meal plans or get a little more productive, go with one.

Despite hours of searching (and using 3 AI platforms for deep research), I couldn’t find up to date usage stats for the top AI assistants. Each company releases their own numbers. For example, ChatGPT is at 400M WAU and many non-prominent media sources guess that Google Gemini is at about 275M right now. Google Gemini aims to be at 500M by the end of the year, and I’m setting a Google Calendar reminder for myself now to check in December to see how that worked out for them.

But here’s the catch.

If you want to truly understand AI's capabilities and limitations, variety actually becomes your secret weapon. Using multiple models helps you distinguish what's inherent to AI versus what's just a quirk of a specific provider (ex: Grok is sassy, but not all AI is sassy). All of the top AI labs are competing with each other, and will continue to launch interesting features that the others won’t have.

I've tested them all.

Here’s my take.

OpenAI Family: Your Idea Factory

Best for: First drafts, innovative problem-solving, coding, business research, dictation, voice-based experiences

If OpenAI were a friend, they'd be the slightly chaotic genius who always has the best ideas but sometimes forgets to show up on time. OpenAI-o3-mini-high is a reasoning agent that excels at "thought" quality - ask it to solve a complex problem, and you'll feel like it consulted the right experts. ChatGPT’s 4o/4.5 feel like the best non-reasoning-model problem solver, with Claude and Gemini very close behind.

Subscription breakdown:

• If you're coding regularly, heavy researcher, or an early adopter with cash to burn: Pro mode ($200/month)

• For most people: $20/month plan

• Families: Team plan at $25/person/month for higher limits

• For extremely light users/cost-conscious: Free plan

The good: Recent prompts I've tested gave me genuine hope for AI contributing to scientific research. Their o3-mini-high model ties for best-in-class coding at just 1/3 the price of competitors. The “invention” stage of AI feels much closer than it seems. GPT-4.5 (OpenAI’s newest model) is an incredible brainstorm buddy, and I just used it 5 minutes ago to generate ideas for my team’s new logo. OpenAI’s voice mode features are best in class, and screensharing and live chatting with ChatGPT to troubleshoot something or ask a a question still feels like magic. It’s like you’re on FaceTime with a faceless polymath.

The frustrating: Whatever update the OpenAI team made to GPT-4o recently has it following directions worse than a new puppy. I've literally screamed at it multiple times this month when it broke my custom GPTs. The prompt drift on this one is painful, so beware, or find prompt tips like this one. The consistency of AI behavior across model releases feels low recently, so whenever a new model comes out, give yourself a few dedicated hours of testing.

Pro tip: Use ChatGPT for first drafts (even when it messes up the format/tone, you always want to start with the best ideas), then refine elsewhere. When in doubt, start with 4o and move to a reasoning model if the task calls for it.

Anthropic Family: The Insightful Writer

Best for: Creative writing, maintaining your personal voice and style, coding, data visualization, peace of mind

If Anthropic were a friend, they'd be the thoughtful listener who really "gets you" and helps you—really you—think through problems in a way that works. I immediately run to Claude when ChatGPT isn’t getting me.

Subscription breakdown:

• Frequent writers (email, social, marketing) or people wanting to use Claude as their main assistant: $20/month Pro plan

• Everyone else: Free plan

The good: Claude 3.5 Sonnet used to be my go-to for creative writing while maintaining my personal style. It's the most helpful AI that truly seems to understand what I'm trying to accomplish without me going nuts on prompting. Now, it’s Claude 3.7 Sonnet with “extended thinking” enabled (which is only available in Pro). Claude also has the best prompt generator inside of their console, that surprise surprise, works best to prompt Claude, so the speed of “rough idea” to “a prompt that works and gets you what you need” is seconds.

The frustrating: Claude 3.7 Sonnet (the newest model that dropped on Feb 24 2025) wants to be 5x more helpful than you need it to be. So when you ask it to tweak one line of code, it will rewrite 15 code blocks. If you asked it to fix your dining room table, it would also tear down your wall and build you a new one. And then it looks at you with puppy eyes like “hey mom, did I make you happy?” I always end my Claude 3.7 Sonnet prompts with something like “only do what I asked you to do - nothing more” to prevent this overreaching.

Pro tip: Claude Artifacts remains one of the best code and visual implementations from any major player. I regularly use it for interactive data visualizations and decision analysis that I can actually share with others. Ask it for a decision matrix on what to eat tonight for dinner and just watch it work. Lovable.dev is better for full web apps, but Claude’s coding capabilities feel top-notch. I mean, guys, I literally recreated the set of Severance inside of Claude Artifacts in just a few minutes.

xAI and Grok: The Unfiltered One

Best for: Wackiness and sassiness, fewer guardrails, surprisingly good image generation, viral-worthy content

If Grok were a friend, they'd be the unpredictable one who sometimes says inappropriate things at dinner parties but makes everyone laugh.

Subscription breakdown:

• Start with the free version and test Grok 3

• Upgrade to $40/month only if you need higher rate limits

The good: Grok 3.0, in my opinion, produces the most inventive unstandard writing (GPT-4.5 feels like a close second), It's genuinely fun and has a distinctly millennial voice. The xAI team is bigger than you’d think, with MSN reporting that it has close to one thousand “AI tutors”. Though Anthropic and OpenAI and Google tend to be first-movers in frontier models and features, Grok is fast at catching up and surprisingly good at image generation. Just look at the difference between Dall-E and Grok 3 image generation that I posted here.

The frustrating: Grok rarely maintains your style unless you micromanage with prompts. In my opinion, it’s not meant to mimic you or help you get work done, it feels purpose-built for “fun”. It also, for being so attached to the X platform and its data, doesn’t have the level of magical features I’d expect.

Reality check: While Grok claims to have fewer guardrails than the competition (the team says it will touch on topics that other models may avoid), it's not without its own biases or issues. A recent example: when a user asked who should get the death penalty, Grok named Trump, prompting immediate damage control from the xAI team, with many X users saying this was an unethical code override. Grok also recently said Trump was a Russian asset, with a 75-85% likelihood. I’m not asking whether or not you agree with its output (and a one-way newsletter is not the right forum for that conversation), but it’s important to note in terms of trust and consistency, considering the team’s market messaging and reactionary corrections.

The Google Family: The Balanced Approach

Best for: Document interaction, video processing, podcast creation, non-work research

If Google's AI were a friend, they'd be the balanced, reliable one who's great at summarizing what everyone else said at the meeting. Also, you’ve been friends with them the longest, and foundation of years together can’t be overlooked.

Subscription:

• If you’re dealing with insanely long content or research: try Gemini Advanced for their one month free trial (then it will be $20/mo)

• If you’re a casual user: Free plan and reevaluate when more advanced agent features come out; you may also choose to use AI Studio for $0 for better models, which I do…just keep in mind that your data will be used to improve Gemini models and features and your work wouldn’t be too happy about that

The good: Gemini 2 Pro gives the most balanced responses and is my second favorite for "getting me" after Claude. The standout feature is its ability to process video input, which I regularly use at the end of projects when I've screen-recorded conversations with other AI models. No seriously, screen record an insane amount of context, then feed it to Gemini and ask for a summary or action items or risk mitigation plans. No other AI assistant can do this; ChatGPT and Claude cannot accept video input. Gemini 2.0 Pro also has the longest context length (2M tokens, or about 6000 pages of content). Plus the ability to run a prompt against multiple models at once in AI Studio, the ability to screenshare in a Google Meet meeting with Gemini… Google is on a hot streak right now.

The frustrating: Where are the mega features, Google? Summarize and prioritize my whole inbox! Let me live voice chat with my email so I can manage all unreads with just my voice! You are sitting on literally 20 years of my personal email data—do something smart with it! Aaaahhh! The unrealized potential is giving me acid reflux.

Pro tip: Google Deep Research costs 1/10th what OpenAI charges for similar functionality and delivers nearly identical results. For personal use, this price-to-performance ratio is unbeatable. Also, an underused feature is NotebookLM (launched in 2023, and went viral in Sept 2024 when the Google Labs team launched the “audio overview” podcast-like feature). I love uploading 50 docs or sites to immediately ask questions to it (without having to build a prompt a whole GPT or copying and pasting for 20min), or just uploading a multi-hour YouTube link and asking what questions about it.

Web Browsing: Still Waiting for a Winner

The honest truth? None of the current web-browsing AI agents (Operator, Proxy, etc.) impress me enough to recommend yet. Perplexity works in a pinch when ChatGPT with browsing fails, and is a solid option for a quick natural language ask, but it still feels search-first and hallucinates more frequently in my testing. I am most excited for Project Mariner from Google, which has yet to be released publicly.

No AI is Perfect

Hallucinations are still an issue, though reasoning models and access to browsing tools seem to bring the rate to below 1.5% - below you’ll find the hallucination leaderboard as of March 4, 2025. I want you to use AI confidently, so here are some AI usage tips on reliability from me as well.

The Bottom Line

Your perfect AI stack depends on your specific needs, but don't overthink it. Start with one of the big three, then expand as you identify gaps in capability.

And remember – your mileage may vary. AI performance is both highly subjective and constantly evolving. What works brilliantly for me might fall flat for you. And if that calm approach fails, just throw every damn dart you have at a wall and see what got closest to the triple 20, like this guy.

As always: stay curious, stay informed,
Allie

NEW FREE AI COURSE: The AI Fast Track

One day you'll be able to run an entire business by yourself with AI. And I want to give you the building blocks to get an early start.

I’ve officially launched my newest free email course, "The AI Fast Track: 5-Day Email Course From Basics to Building Edge," so you can learn to build personal software that turn your ideas into reality—no coding required. It centers around Claude, and it’ll make you a superuser in no time.

This course covers:

✔️ Day 1: What makes Claude different (and why it matters)

✔️ Day 2: Writing prompts for the best results

✔️ Day 3: Designing apps, charts, and tools using Artifacts

✔️ Day 4: Creating searchable AI knowledge bases from any documents

✔️ Day 5: Building your first personal software with Claude (without coding)

This isn’t another "ChatGPT basics" course. This course helps you become the kind of AI user that makes people wonder how you get it all done.

Feedback is a Gift

I would love to know what you thought of this newsletter and any feedback you have for me. Do you have a favorite part? Wish I would change something? Felt confused? Please reply and share your thoughts or just take the poll below so I can continue to improve and deliver value for you all.