Apologies for the delayed release on this one – I know you all were waiting with bated breath and I dramatically overplayed my hand last time when I promised this update “next week”.

My rationale for why it was cool for me to keep you waiting? My daily AI drivers have changed multiple times over the last few weeks, so it made sense to try out the new toys. Also, I tore my ACL & had to have surgery, both kids have been sick, ***load of work, etc. You know what they say about excuses right?

Anyway just like Marc Benioff, I’ve been using ChatGPT basically all-day, everyday for 3 years and now I’ve mostly ditched GPT for Gemini & Claude. Luckily for OpenAI I have kept my max tier sub because I invested in creating a ton of Custom GPTs. And their Codex CLI product still feels like the most reliable coding agent, which I have pretty extensively customized for several of my current coding projects. But I’m not sure how long OpenAI can keep me with how fast the others are gaining ground & providing more bang for the buck.

Everyone is talking about how Gemini has come out strong this AI season, and I can’t help but agree. They’re offering a really compelling AI product suite that is starting to integrate seamlessly with all these Google products people already use. And in ways that are actually useful. I’m still just scratching the surface myself, but Gemini has integrated with Google Photos, Google Earth, Google Maps, Google Home – all us lowly Android users get Gemini as our assistant…the list goes on. You don’t have to be an industry insider to see that Google’s position as an already-profitable company with many revenue channels seems far sturdier than OpenAI, who basically has the-currently-unprofitable-ChatGPT…and nothing else. But the same could be said about Anthropic, and both of these companies seem to have no problem finding a seemingly infinite, voracious capital stream to feed the beast that seems driven more by FOMO than PE.

From OpenAI’s perspective, Anthropic’s Claude should be viewed as just as potent a competitor as Gemini. The recent Opus 4.5 update is a serious upgrade to previous models that puts it on par with Gemini for reasoning and arguably better at certain tasks involving code or finance. Claude also gives me by far the best results for productivity tasks like editing documents, because its Claude Code DNA allows it a far greater degree of precision than the other major LLMs. Anthropic has been positioning itself as the ‘AI for business’ and you can really feel that when you’re using the product.

So without further ado, here’s my personal ‘power rankings’ for my daily drivers:

RossAI Power Rankings*

General Use	Video	Photo
Gemini 3.0 – now the best at most things. Best reasoning, very fast, gigantic context window. Lots of additional features – I use the Gemini Live voice-to-voice a lot.	Google Veo3 – best EZ button video generator right now. Incredibly lifelike humans, but requires detailed prompting and patience.	Gemini nano banana – delivers incredible photorealism, complex scenes, continuity. It’s crazy how far this has come.
Claude Opus 4.5 – the best at precision tasks or specifically business / finance, or coding tasks. Opus is actually pretty incredible, the only downside is you get less bang for your buck here vs the entire Gemini AI suite.	Kling AI, Higgsfield & Weavy AI – when you need long-form AI video, Kling, Higgsfield and Weavy are the way. Require a little more knowledge but allows you to get incredible specific results.	Midjourney – when you need something more vivid / fantastic / creative, MJ is the way. Can take a little more tuning than NB and might take more tries to get something that looks ‘real’ if that’s what you’re going for.
Chat GPT – now pretty much only if I’ve already built a custom GPT or have a detailed folder for that task. Its 5.2 update only just came out so we’ll see how it compares to the massive upgrades Google & Anthropic just put out.	OpenAI Sora – 2nd best EZ button video generator. Fun horserace between Sora and Veo, but right now I think Veo is slightly ahead. This seems to literally change monthly so plan on following them both if you’re in this biz.	Adobe Photoshop – if you want to edit existing photos using AI, photoshop’s capabilities are now at the near-magical level. It is truly insane how easy it is to do things I used to describe to clients as impossibly-expensive.
Grok, Meta, Deepseek, etc – I honestly rarely even boot these other than out of curiosity to see how their updates are coming. Grok/Meta are fast on processing stuff if you want a large data table formatted. But other LLMs do pretty much everything better rn.	Stable Diffusion – what a lot of pro’s and internet memes are using to create crazy AI videos with real world figures, violence, etc…because it has no content filters. Open source & runs on user machines.	Stable Diffusion – comes in both image and video flavors. I don’t find myself needing to use this often, but if content filters – which are aggressive – are tripping you up on a project, this is where I’d go next.

Design	Coding	Organization
Adobe Firefly / CS AI tools – allow you to bring generative AI into your existing power suite which lets you actually use a high degree of precision with your generative AI. Of all the things, AI still kinda sucks at some types of design – looking at you logos – so most designers will want to use it for some mechanical elements but not everything.	OpenAI Codex / Claude Code. Huge battle between Claude Code and Codex. I’ve switched multiple times back & forth between them. The consensus seems to be – as of today – that Opus 4.5 is smarter / more elegant, but GPT-5.2 Codex might be more reliable at implementation.	Notion 👑- I’ll write a whole article at some point about how much I love Notion. Notion is how I keep everything I do straight & I also use its platform AI features constantly because it lets you edit datasets super easily.
Canva – stepping up the ability to do layered generative similar in theory to Adobe, and its much easier to work with.	Cursor, Windsurf, Loveable – make vibe coding a little more accessible with a nice web UI wrapper, allow you to easily swap between models. But generally speaking you do pay more per task to use them vs just getting your code agent’s straight from the source.	NotebookLM – google product that’s well integrated w/ Gemini and might work better for some people if they want a simpler system than Notion.
Figma – generative UI/UX features can feel like magic for web / app designers
Claude is actually great at web design / UI prototypes.

*as of 12/19/25

Honorable mention / tool that doesn’t quite fit a category is ElevenLabs. ElevenLabs does SOTA voice generation, which has been incredibly impactful in several of the businesses I’m working on. It allows you to create extremely realistic voice clones that can read scripted text, and also allows you to very easily create voice agents that can access and leverage a custom knowledgebase you provide to do some pretty serious heavy lifting in terms of conversational engagement with customers / suppliers / etc. I have used it quite a lot for several production projects and I think it’s one of the stronger AI platforms around.

These are totally unofficial and probably will be completely outdated starting tomorrow. There are however a few places that actually keep very on-top of this stuff that you can bookmark if you want to be really informed. It will make you super popular at all the parties I promise, everyone loves talking about this almost as much as they love hearing about my fantasy football team.

LMArena:

Overview Leaderboard

Artificial Analysis

https://artificialanalysis.ai/leaderboards/models

Vellum

https://www.vellum.ai/llm-leaderboard

LLM Stats

AI Leaderboards 2025 – Compare All AI Models

I probably missed some good ones – let me know in the comments if there are some others I should have included.

Final thoughts for this one – I think most of the world doesn’t realize how many LLMs there really are. While my list above is basically all ‘frontier’ models that are closed-source and ‘best of the best’, there are now thousands of open-source LLMs and SLMs (small-language models), many of which are completely free if you have the hardware to run them.

Check out https://huggingface.co/ for a gallery of some of the more popular ones.

A lot of these open-source models can do basic processing work as well or better than frontier models, and can be run on a pretty standard desktop computer.

The margin between these models and the frontier models broadly seems to be shrinking more than it’s growing. In many cases, people are starting to realize that highly specific, task-oriented small-language models (SLMs) are way more efficient when it comes to tasks like data processing and computation than LLMs. The analogy that stuck with me was “using an LLM for basic math is like asking an English PhD to do algebra.” He can probably do it, but you don’t need to know Shakespeare to populate my spreadsheet.

For the most part, I still gravitate towards the big guys, but I have definitely been playing with Ollama and some math-specific models on some of the AI finance projects I’m working on.

Long story short, as we all know, it’s a fast moving world to try to keep up with.

What I’ve learned is basically just to budget time for consistently researching where the bleeding edge is, and to try to prepare for switching costs because I know I’m going to change tools constantly. Now, when I’m building custom GPTs or Gems, I try to put all my source code / instructions into Notion first to keep everything platform agnostic.

Alright that’s what I’ve got for today. Let me know if there are AI tools you’re using that I missed, or if you think I’m completely wrong on my takes.

To keep myself honest I’ll say there’s probably no way I’m putting up another article next week, but in the next few weeks I’ll organize and share a database with a bunch of the AI accounts I follow on various platforms to stay informed.

Cheers and happy holidays!