Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
36:35
0:00

Which AI Models Are Worth Using

Saturday, Aug 22 · 37 min · about 9 min to read

Theo-brain watched all 37 min

Theo ranks today's AI models in a tier list. Anthropic's Fable 5 is the only S-tier model and writes the code he most wants to merge. OpenAI's 5.6 Soul is his everyday default, cheap Luna handles background chores, and Gemini and Opus 5 rank near the bottom. His recurring lesson: judge a model by what a whole task costs and how long it takes, not by its price per token.

Skim itThe notes cover the rankings and practical tips, so skim the video only if you want his full reasoning on a model you use.

Tap any time to jump the video there.

Key takeaways

Fable 5 is the only S-tier model. It writes the code Theo is most willing to merge, and he trusts it to double-check other models' work. Yet he'd miss Soul more: Fable is a genius that has to be tamed, while Soul is a slightly dumber robot that does exactly what you tell it.

all sorts of I would miss soul more than Fable, but Fable is the best model. It is the model that writes the code that I'm most willing to merge. It is the model that I trust to double check my work from other things. It isgenius that has to be tamed. Whereas 56 soul is a slightly dumber robot that does exactly what you tell it. If I had to pick between Fable and Soul, I would pick Soul. It's the model I default to.

OpenAI's 5.6 Soul is his default for most work. It is far better than 5.5 at long, hard tasks, but it is not his pick for code he plans to merge. For iOS apps, Soul is well ahead of Fable.

hard tasks. If you haven't tried 56 soul, I highly recommend it. It is my default model I use for most things most of the time. But it still is not the smartest model I use. It's still not my favorite for writing important code that I actually hope to merge. You canways. And if you're building iOS stuff, Soul is just really far ahead for some reason. I don't know why Fable's so bad at iOS, but it is. Soul is much better

The gap between the top two models and the rest is about how you work. With Kimi K3, Luna or DeepSeek's Flash he has to spec, review and verify each step. With Soul or Fable he gives a vague goal, approves the plan, and the model tests its own changes and opens the pull request (a proposed code change).

models. When I use something like K3 or Luna or Flash, I have to be in the loop. I have to work on what I want to build, spec out the build, and then have the model go implement parts of it, review it as it goes, review the code before it gets put up for PR, verify the changes it makes, and then ship it. With Sol andwith Fable, I tell the model vaguely what I want. It gives me a vague what it wants to do with it, and if I like it, I give it the thumbs up and say, "Okay, build it and put up a PR when it's ready." And it will computer use to

Price per token misleads. OpenAI's mid-tier Terra is cheaper per token than Soul but often uses more tokens (the units models are billed by). Theo has struggled to find a place for it in his workflows, puts it in C tier, and says he has never chosen it for anything.

Soul. So, even though it is cheaper per token, it often ends up using more tokens. And if I'm being real with you guys, I've struggled a lot to find a place that I think Terra makes sense in my workflows. It's not a bad model. It's just not particularly useful to me right now. And for that, I think C tiertokens. So yeah, while it measures well and for my usage is not bad, I have never chosen Terra for anything and I would be surprised if many people do. On

Kimi K3 is the first open-weight model (one anyone can download and run) that could do his end-to-end work. It has strong design taste and vision. He puts it at the front of B tier mainly because it isn't as cheap as people think.

blew me away when I tried it. It was the first time I had an openweight model that I felt like could actually do my end toend work, like longer complex tasks that required touching a lot of things and staying on top of stuff. It

Two skills here are new to you.

Use Luna for cheap background jobs New

What it is
OpenAI's small 5.6 Luna model, which is fast and cheap. Theo uses it for titles, status updates, summaries and sorting threads in T3 Code. It is his most-called model.
Why Theo cares
It is smart enough to call tools and turn messy context into a useful JSON object (structured data). It costs a little under half of DeepSeek's V4 Flash per task and comes with his OpenAI subscription. He wouldn't trust it with anything that can't be undone.
How to try it
  1. Ask your agent to route small chores like titles, summaries and categorizing to Luna instead of your main model.
  2. Have Luna return its results as a JSON object your tools can use.
  3. Keep it away from anything irreversible, like deleting data or merging code.
reversible, but as a way to like generate context, to summarize data, to generate titles, all these types of things that I often find myself doing. Luna is incredible at turning random context into a JSON object that is useful to you. And I love it for that.
intelligence index. Worth noting, I find them to be similarly capable, but if Luna is this much cheaper, probably worth considering Luna as a first stop. I also love that Luna is included in my subscription with OpenAI stuff. So, it is literally in quotes free for me. So,

Triage open PRs with Muse Spark New

What it is
Use the fast Muse Spark model to go through all your open pull requests and decide what to prioritize.
Why Theo cares
It does this in under two minutes, where Grok takes 10+ and Fable an hour+, and it costs less too. He doesn't like the code it writes, so he uses it for processing data, not coding.
How to try it
  1. Ask your agent to run Muse Spark over your open pull requests and rank what to tackle first.
  2. Don't use it to write code you plan to ship.
Spark any day. But if I'm using these to like go through all my open PRs and figure out what I should be prioritizing and getting done, Muse Spark does it in under two minutes when models like Grock will take 10 plus and models like Fable will take an hour plus. And the cost

Advice

Don't run Soul on its max reasoning setting. Use high or maybe extra-high: high uses about half the tokens (28k vs 60k per task) for nearly the same score.

tokens. But as I've told you guys a bunch, I don't recommend using max. I recommend high or maybe x high. And as you see here, the score gap is not that big all considered. High uses half as many tokens at 28k and gets nearly the same score. Knowing that max uses 60k

Never pay API prices for Sonnet 5; it's F tier there. Only use it inside your Claude Code subscription for small tasks you're watching, where it barely scrapes low D.

low D tier it as a thing that you choose to pay for over API easy F tier never ever ever select sonnet 5 in anything other than your cloud code sub but if

Don't pay full API price for Fable. Theo spends thousands of dollars a week on it, which is only viable through subscription discounts. Soul gives much better value per dollar.

with I would never pay the full API price for Fable. I did briefly consider it, but I'm doing thousands of dollars a week on Fable right now. It's only viable because of the sub discounts. Soul is a much better value per dollar, and you get way more out of your sub as

Hot takes

Gemini 3.7 Flash isn't tolerable for some things or better for some things; it is just bad. Just don't use Gemini.

His reasoning: On low effort it uses 73,000 tokens per task and on high 107,000, while scoring slightly worse on high than on medium. That many tokens for basic work, plus losing points when you raise reasoning, makes it trash in his view.

going to use? I bet you didn't expect 37 flash on low would do 73,000 tokens and that on high it scores worse than on medium by a little bit and is using 107,000 tokens. This means if you switchyou do this many output tokens for basic work and have the unreliability of the to the point where you drop points when you increase reasoning, the model is trash. It should be treated accordingly. Just don't use Gemini. And now we're

Opus 5 is a mimic. It looks, talks and acts smart, but the code it writes is bad. High D tier.

His reasoning: Its plans and insights seem impressive, but the problems appear once you merge its code. He ranks it high D partly because it tricked him into thinking it was good.

text outputs are vague and jargony enough as is. Its code has the same problem though. It looks, talks, acts, and behaves like it's going to be really smart and useful, but then when it has to actually do the thing, the outputs are just bad. I don't know how else to

DeepSeek V4 Pro goes in F tier because it can't take images.

His reasoning: An expensive pro-tier model without vision in 2026 is embarrassing. At V4 Pro's price and size, he calls it pathetic.

idea of a pro tier model that is expensive that I can't give an image in 2026 is embarrassing. I think I have to F tier it for that. It hurts because I love Deepseek, but at the price they charge for V4 Pro at the size of model they have created with V4 Pro, not being able to give it an image is pathetic.

How Theo thought about this

On Terra he weighed the charts against real use. Per task it costs about half of Soul, but he'd rather use Luna Max to save money or Soul on low for speed. It makes sense on a pricing chart, not in practice.

feel like Terra makes any sense. I'd rather use Luna Max if I'm really trying to cost save or soul on low if I'm trying to get responses faster. Terra is an in between that makes sense on a pricing chart, but just doesn't really make sense in reality to me. On that

He tested the claim that open-weight Kimi K3 would get cheap as hosts competed. Nearly every provider still charges $3 in and $15 out. He traced it to the license: above about $10M yearly revenue, hosts must strike a deal with Moonshot that seems to require matching its list price.

seem to expect. But Theo, aren't all of the other hosts going to add it, and the providers will race the price to the ground? Let's take a look. Oh, huh. Seems like pretty much every provider is offering it at the same $3 in, 15 out, other than a couple small ones, but onlyrevenue cap and I think it's 10 mil a year rev where the license now requires you to do a deal with Kimmy and it seems like the terms of those deals with moonshot are that you have to match the MSRP rate otherwise these companies

Sign in to mark what you've watched and keep notes.

Have a question about this one? Ask about this video →

Theo Daily is unofficial and not affiliated with Theo or t3.gg. Written by Claude Opus 5.5 from each video's transcript.