Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
15:50
0:00

OpenAI is getting desperate

Monday, Nov 2, 16 min, about 6 min to read

Theo-brain watched all 16 min

OpenAI's doubled limits are real on the $100 and $500 plans, but the $200 tier merely gets back last month's cut.  At $200, Theo estimates about 205 monthly code reviews with Opus on Claude versus roughly 290 with Sol on ChatGPT Pro, a narrow gap.  Quality decides it: in blind tests even Sol preferred Opus's reviews 54 of 83 times, so Claude remains his pick for daily coding. 

Skim itWatch the usage-limit chart to see how the resets change the apparent improvement in OpenAI’s limits.

Tap any time to jump the video there. Quotes are YouTube's automatic captions, word for word, so some names come out misspelled.

Key takeaways

The doubling is real for the $100 and $500 plans. For the $200 plan it is only a restoration: the plan is back where it was a month ago, not a real bump.

It is worth calling out that the $200 plan is back where it was a month ago. It's not a huge bump for $200 subs. It's just a restoration of where they were.

In real use the gap is small. On a $200 plan, Theo estimates about 205 code reviews a month with Opus on Claude versus about 290 with Sol on ChatGPT Pro. The prices look far apart, but the usage is close.

…on the ChatGPT Pro $200 plan, it comes out to a much closer gap here with Opus giving you around 205 reviews on that plan and the ChatGPT Pro sub giving you 290 reviews.

Opus wins on quality. In blind comparisons of code reviews, Opus picked the Opus review 66 of 83 times. Even Sol preferred the Opus review 54 of 83 times.

Opus preferred Opus 66 out of the 83 times. Soul preferred Opus 54 out of the 83 times. So, Soul preferred Opus 65% of the time, Opus preferred Opus 80% of the time.

Measured in raw dollars of usage, Claude's $200 plan still gives more. Even with half limits on Anthropic's Fable model it is worth about $6,000, against about $4,300 on OpenAI's $200 plan.

But even then, the $200 Claude plan with the 50% cut gets you six grand of usage, where the $200 plan on Open AI, no matter what, only gets you 4.3k.

Two skills here are new to you.

Council: multi-model review New

What it is
A skill Theo uses where three to five different model families each review the same change, plan or proposal and give feedback.
Why Theo cares
He gets feedback from several models at once. Because every model gets the exact same task, it also doubles as a fair like-for-like test of which model reviews best.
How to try it
  1. Before approving a plan, ask your agent to send it to three to five different model families for review.
  2. Compare their feedback side by side on the same task.
  3. Keep a running record of which model's reviews you end up trusting.
…some changes I want to make, I have a proposal, a plan, or something. And then the Council comes in across three to five different model families to give feedback.

Theo's subscription split New

What it is
Theo splits his AI subscriptions by job: his OpenAI subs for computer use and verification, his Claude subs for all his day-to-day coding, and one Grok sub just for its Grok bot assistant.
Why Theo cares
Each provider covers its own job. He has mostly been using Grok bot and calls it surprisingly solid. Elsewhere in the video he calls Anthropic's Opus 5.5 the best model available.
How to try it
  1. Decide which job each AI provider's subscription is for instead of using one for everything.
  2. Keep a separate subscription only for a job it clearly does well.
I did get invited. I've mostly just been using Grok about though. It's been surprisingly solid. It's kind of crazy that I have my OpenAI subs for computer use and verification. I have my Claude subs for all my day-to-day code work.
And if you doubt that, I'm sorry. Opus 5.5 is the best model available, at least as of the time of this recording.

Advice

Don't run models on their Max reasoning setting, the most thinking per answer. Theo almost never recommends it on any model line. He mostly uses High and occasionally switches to the step above it, XHigh.

But remember, I almost never recommend Opus 5.5 Max. I basically never recommend Max on any of the model lines, especially in profit models. I mostly rock high with the occasional shift to XI.

Choose ChatGPT Pro only if getting 30 to 50% more work done each month matters more to you than the weaker capability of OpenAI's Sol model. Otherwise, take the Claude plan.

So, if a 30 to 50% bump in things you can do in a given month is super, super important to you and you're willing to take the hit with 61 Soul's capability, then do it.

Hot takes

OpenAI's real problem isn't limits but its models. It has nothing that can go head-to-head with Opus 5.5.

His reasoning: He calls Opus 5.5 unbelievable and says nothing OpenAI has released is quite there yet.

OpenAI, I'm begging you. We need something better. We need something that actually can go head-to-head with Opus 5.5 because that model is just unbelievable and nothing…

Comparing models by price per token is misleading.

His reasoning: Models use very different numbers of tokens (chunks of text) to solve the same problem. A model that is cheap per token can still cost more per task.

…problem, which is the thing I think most people miss is they look at the dollar price per token and they forget that certain models use more and fewer tokens.

The outrage over ChatGPT and Work chats now counting against Codex limits doesn't make sense.

His reasoning: Chat use burns far fewer tokens than daily Codex coding, and the doubled limits should cover both.

I don't think people being outraged about that are making much sense. Usage in work and chat is obviously going to use exponentially fewer tokens than day-to-day Codex usage.

How Theo thought about this

Theo caught himself mid-analysis. One chart looked much better for OpenAI, and he first guessed it included the doubled limits. It didn't: it counted OpenAI's frequent limit resets, 11 in the last 30 days. Without the resets, OpenAI's limits look much worse. He checked what drove the numbers before trusting them.

I thought the reason was because they included the 2x limits here and they might not have in the other chart I was in. That is not the case. The reason is much sillier. This version has resets because if you didn't know this, OpenAI resets a lot, 11 times in the last 30 days.

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.