Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
48:21
0:00

Nobody Is Competing With Anthropic

Tuesday, Nov 10 · 48 min · about 10 min to read

Theo-brain watched all 48 min

Anthropic released Fable 5.2 and Mythos 5.2, the same model behind different access rules. It tops Anthropic's benchmarks and is cheaper and faster, but in Theo's own tests on his real codebases it trails Opus 5.5 and needs more hand-holding. His conclusion: keep Opus 5.5 as the daily driver and bring Fable in for hard, open-ended problems.

Skim itThe notes cover the arguments; watch only for the visual demos and Theo's closing comparison of the two models.

Parts

The announcement: Fable vs Mythos

Fable and Mythos are one model. Mythos simply has fewer outside filters blocking requests and is limited to vetted cyber and bio groups.

but it's the same building. The models are the same. Mythos just is less likely to have a third party around it, block the request or the response. Fable has

Benchmarks: Anthropic's vs Theo's own

Fable 5.2 dominates Anthropic's benchmark suite, but does noticeably worse on the benchmarks Theo built himself.

This is one of the most utter dominations of a benchmark suite I've seen in a while. But don't worry, there are other benchmarks, many of which I've made. And it does not do quite as well

Anthropic's coding and science claims

A Python-to-Rust migration and an eight-day prototype OS no longer impress Theo, since earlier models already did similar work.

Cool. Good work. Somehow not that crazy impressive to me anymore. It's it's insane how fast we got used to this. Like the this frog got boiled fast. Here

Why enterprises flocked to Opus 5.5

Opus 5.5 kept companies' zero-data-retention terms intact, which drove huge business adoption.

that a lot of enterprises could use with their existing policies without losing ZDR, which resulted in just world-breaking adoption of Opus 55 at a lot of businesses. It is hard to put

Theo's own evals and code review

On his own repos, Fable makes some false claims. It is a careful code reviewer but a very slow one.

And in the code review stuff, I found it to be quite good overall. It does take far too long to do these reviews, which honestly kills a lot of the value for

Subscription limits and pricing

Fable usage counts against a separate, half-weight limit on the $200 plan. Theo tests what a Fable-only reset gives back.

There's an issue though. This is a model reset. What does that mean? It means it resets your limit with Fable. Not your overall limit, just Fable. But

Design, quirks and model comparisons

Real work examples, good and bad. Overall Fable is pleasant and does what it's told, but it makes odd mistakes.

is pathetic. What I'm trying to say is this model is pleasant to work with. It does what you tell it and it does it

Fish Slop game test

Fable's 3D fish game looks and runs better than anything Theo has gotten from OpenAI models.

This runs a hell of a lot better than anything I've gotten out of an OpenAI model, and it looks better overall, too.

Verdict: workhorse vs clever friend

Opus 5.5 does exactly what it's told. Fable tries too hard to be clever but excels on the right hard problem.

small, it's not that it's dumb. Opus 5.5 is just a workhorse. It does exactly what it is told to do, whereas Fable tries a little too hard to be clever.

Tap any time to jump the video there.

Key takeaways

Opus 5.5 is still the model to use by default. Theo once ran a project audit on Opus by mistake, thinking it was Fable, and was impressed by the result. Opus also stretches your usage limits and API budget much further.

switch to Fable because Opus was my default and I initially ran on the wrong machine, so I swapped machines and then didn't switch the model and didn't notice until after I read it and was impressed. This is going to be a commontheme. Opus 55 is still insanely good and you can get way more done with the limits you have available to you and if you're actually paying API prices, even more than that. Opus 55 is a gift

Fable 5.2 wins Anthropic's benchmarks, but on Theo's own tests it does worse. He has models analyze his real codebases and review real pull requests (proposed code changes). There it trails Opus and Sonnet in almost every test, though it still beats OpenAI's models.

stuff like that. And I found that Fable 5 2 is actually underperforming Opus and Sonnet and massively outperforming OpenAI models in this particular eval set.

Fable is more capable but not more autonomous. It needs a human in the loop more than Opus did: it can latch onto an unimportant decision and spiral. A friend's framing, which Theo endorses: it is better at not getting blocked on hard problems, not at making judgment calls alone.

needs it a bit more than Opus did. Like it needs you to hand hold it a little bit or it can get itself stuck pretty quick. Not stuck as in like it can'ttell, I didn't. This model is much more capable, but it does not have much more autonomy. It's not better at like making decisions by itself in long runs. It's better at not getting blocked in long runs and solving hard problems in long

On the $200 plan, Fable has its own weekly limit on top of the regular one. Every 2 points of Fable use counts as 1 point of the regular limit. If you max out Fable, your monthly value drops by about $4,000. Used alone, Fable's value falls from 73x what you pay to 15x.

The Fable limit is an additional limit on top. Every two points you burn in the Fable 5 limit counts as one point in the regular limit. So you can get the Fable limit down to zero and still have 50% left in the week. This is the state I55. So you end up taking a $4,000 hit on the total you can spend in a month if you max out your Fable usage. Could be worse. Obviously that still sucks and if you're only using Fable your multiplier goes from 73x to 15x. It's an 80% decrease.

Fable 5.2 is cheaper than before: $8 per million input tokens and $40 per million output. A new optional fast mode costs twice as much for 2.5x the speed. Theo calls that a fair deal, better than OpenAI's fast modes, where the price jump isn't worth it.

also cheaper too. $8 per million input tokens and $40 per million out. Fableat their stock speeds. You had no way to go faster. Now you can pay twice as much money in order to get it 2.5 times faster. Not actually that bad, especially considering how much less a bump normal fast mode is from OpenAI.

Two skills here are new to you.

Anthropic's design skill for front-end work

What it is
A skill is a short markdown instruction file the model loads. Anthropic provides one for interface design.
Why Theo cares
With the skill on, Fable's page designs had real ideas and nice animations. With it off, he got generic 'Tailwind template slop' and the overused brutalist look. A little markdown makes a big difference.
How to try it
  1. Ask your agent to turn on Anthropic's design skill before any UI or landing-page work.
  2. Generate the same page with the skill on and off and compare them yourself.
skill or you're going to end up with fable slop. Claude slop in general in the UI is a thing you have to beat out of it and thankfully they provide that skill. It's crazy how a little bit of markdown can help, but it does a lot.

Fix a model's habits through AGENTS.md New

What it is
When a model keeps doing something you dislike, ask it what that pattern is called and why it does it. Then add a rule to your agent instructions file (agent MD).
Why Theo cares
Fable kept adding needless 'eyebrow' text and subheadings. After he asked it about them and updated his agent file, it did them much less.
How to try it
  1. Ask the model what it calls the unwanted habit and why it does it.
  2. Have your agent write a rule against it into AGENTS.md.
  3. Check over the next few tasks whether the habit returns.
lot of my stuff too, so I asked what it calls it and why it does it. It told me what to do, so I made some changes to my agent MD to make it not do that and it hasn't done them as much since, which has been nice. It also seems better at

Fable as a careful pull-request reviewer New

What it is
Use Fable 5.2 to review batches of PRs, the code changes waiting to be merged.
Why Theo cares
If Fable says a PR is good to go, it usually is. It also skips the minor nitpicks other models get stuck on, though it misses some things. The catch is speed: about 120 minutes for 8 PRs, versus roughly 20–30 for GPT models.
How to try it
  1. Run Fable as a final check on PRs that aren't time-sensitive.
  2. Use a faster model when you need a review back quickly.
the very least. Generally speaking, I've had pretty good luck with if Fable thinks a PR is good to go, it'd be surprising if it isn't. There's a
me. The fact that it took 120 minutes or so of runtime in order to review eight PRs is rough. Yeah, too much time for not quite enough of a win here. I don't have the numbers logged for the GPT models here, but they took a lot closer to like 20 to 30 minutes for this batch.

Sol 6.1 for codebase analysis

What it is
The model Theo calls a workhorse for analysis work, such as explaining a large codebase.
Why Theo cares
Analyzing his TypeScript-in-Rust codebase, Sol 6.1 made 38 claims with none false, in 29 minutes, for $4.19. Fable made 61 claims with 4 wrong, at about $23.
How to try it
  1. For 'explain or audit this codebase' tasks, try Sol 6.1 first.
  2. Spot-check a few of its claims against the code.
Fable 52 still makes wrong claims here and there. Of the 61 claims it made about code base four were wrong, but only cost around $23 to do this thorough analysis of TS-Rust. Remember, complete
second place behind 61 Soul. 61 Soul made 38 claims, none of which were false. Did it in 29 minutes for $4.19. Saying this for a bit, 61 Soul is a workhorse for analysis type stuff.

Advice

Don't use Max reasoning effort for coding. It costs far more and tends to overthink into mistakes. Anthropic's own benchmarks now show Fable 5.2 Max scoring worse on most coding tasks.

reasoning effort has always been an iffy balance for code where not only is it much more expensive, it often will go too far and make mistakes as a result. This is now encoded in the benchmarks. Fable 52 Max will score worse on most coding tasks consistently and you should not use it as a developer. If you think

Keep auditing what the model produces. Fable built Theo a 'safe mode' banner meant to hide his email during recordings, then displayed his email inside that banner. Opus would not have made that kind of mistake.

that nothing leaks and then it just leaks a bunch of things in that bar. It's so stupid. Just the types of mistakes I would never have expected even Opus to do, this model will sometimes do. You still got to audit your outputs. I like

Give Fable the source material, such as your past chat history, so it can check its own claims. It will admit when it's wrong, but it also isn't afraid to push back.

confirmed I was right. You give it the sources for the data, it can usually disprove itself, but it's not scared to push back, which are the things I really like about it. But it also just misses

Hot takes

Fable 5.2 is almost certainly a smaller or cheaper-to-run model than Fable 5.1.

His reasoning: Its speed jumped from 81–84 to 106 tokens per second, close to Opus's 112. It is clearly cheaper for Anthropic to run, which explains the price cut and the new fast mode.

bit slower. It was 81 to 84 TPS with Fable 51 and this one's 106 TPS versus Opus at 112. They are spitting distance from each now. This model is not slow, which means we're in conspiracy land. This is a smaller model.smaller or something. It is clearly cheaper for them to run. That is why they are doing what they are doing. That's why they're offering fast mode now. That's why they made it cheaper. That's why it is faster objectively.

The subscriptions aren't simply subsidized; API prices are inflated.

His reasoning: Anthropic's margins on API prices are so high that enterprises effectively pay a tax that funds cheap subscriptions.

inflated. Everyone looks at this as the crazy subsidization of the subscriptions and it is that. But on the other hand, the margins that they charge are so insane on the API prices that they get away with it. Think of it as some part us paying less on the sub, some part enterprises paying a huge tax on top of

Fable 5.2 won't become his model for everything, the way Opus 4.5 and 5.5 each did. That's why it's called 5.2, not 5.5.

His reasoning: The jump from Opus 5.5 to Fable 5.2 is much smaller than the jump from Sonnet 5.5 to Opus 5.5. If it were good enough to be 5.5, it would have been named that.

the model I used for everything. Fable 5.2 just dropped, and it is not the model I'll be using for everything. I am hopeful this is meant to be a signal toif the gap between Opus and Fable was bigger, but the gap between Sonnet 5 5 and Opus 5 5 is significantly bigger than the gap from Opus 5 5 to Fable 5 2. The gap was the same. They would have maintained the numbers. Gap isn't, so

Anthropic's claim that over 90% of its merged code is written by Claude now sounds modest.

His reasoning: What people mocked six months ago is now true at many companies. He has heard from Boris at Anthropic that the other 10% is also Claude, just not properly attributed.

what it is worth, I have heard from Boris in some interviews and other places that the other 10% isn't not Claude. It's not properly attributed, and they'd rather give the low numberwith Claude. The things that they said 6 months ago that we all made fun of them for are now true for many businesses outside of Anthropic, and now their claims feel reserved, which is really funny. Team of agents running an earlier

How Theo thought about this

Nobody could tell him whether a Fable-only limit reset also restores half your weekly limit. So he spent one of his valuable resets on camera to find out. It did.

I want an answer. So we're going to figure it out together. Will it give me 50% back on my week? Cuz that's what it should do. Here it shows for the week I have 51% used and for Fable I have 0 percent. So it does indeed give you back half your weekly limit. Cool to see. I

In the Fish Slop game test, Fable's version looked better, but Opus's gameplay felt better. He played the Opus game to completion because its gameplay loop was fun. He concluded that a combination of Opus and Fable is the way to go.

The gameplay is still not as compelling as it was with Opus, and once again, I feel like a combination of Opus and Fable is the way to go.But I also found the gameplay felt better in this one overall, and I actually ended up playing this one to completion because I found the gameplay loop to be actually quite fun. So yeah, take that as you will. I find that Opus actually makes more fun games.

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.