Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
30:58
0:00

OpenAI fights back

Tuesday, Sep 29 · 31 min · about 7 min to read

Theo-brain watched all 31 min

OpenAI's GPT-6.1 Sol is a cheap new model that Theo finds far more reliable than OpenAI's erratic Astra and excellent at digging bugs out of code. It is weak at long unattended building and bad at UI design. So Theo keeps Anthropic's Opus 5.5 as his main builder and brings in Sol as its reviewer and investigator.

Skim itThe notes cover the substance; watch the game demo if you want to see the design problems for yourself, and the code review story if you want the detail.

Tap any time to jump the video there.

Key takeaways

Sol costs $2 per million input tokens and $10 per million output (tokens are the chunks of text models bill by). Cache reads, text the model has already seen, cost 10 cents per million: a 95% discount instead of OpenAI's usual 90%. That makes real-world agent work much cheaper.

this model comes in. This model is $2 per million input tokens and $10 per million out. This makes it way cheapergoing to be 10 cents per mill in. That's a big deal. OpenAI has not changed cash read price as far as I know ever before. It's always been exactly 10% of the normal read price. That makes it a 90% discount, and now it's a 95% discount. That means they cut the cash read cost in half, massively reducing the cost for

Sol fixes most of Astra's 'spikiness', its habit of suddenly doing something very dumb. Theo had Sol go through his email for overdue invoices and prepare the bank wires for him to send. It got nothing wrong and flagged things he would have missed. He would not have trusted Astra with that.

leave me to hit send. It didn't get a single thing wrong and called out additional stuff that I absolutely would have missed if I was doing this work myself. So, I'm literally trusting this model to wire money for me. It's trustworthy enough for that, and

Sol's standout strength is review. On a real bug-fix pull request (a proposed code change) in his app T3 Code, it found real problems that both Fable and Opus had missed. That kept a regression, a fix that breaks something else, out of the codebase.

just a placeholder. So, when I had 6.1 Soul look through this, it found real problems that were entirely missed by Fable and by Opus. First, it called out

For long, unattended rebuilds, Sol is weak. On Theo's TypeScript-to-Rust port, it burned lots of tokens and made no progress. Opus got the port working in about a night for around $1,000, on two Claude subscriptions. His verdict: Anthropic is 'comically far ahead' here.

did it in like a grand in like a night with just two subscriptions with the Claude plan. So, for unattended long like heavy rewrite type stuff, Anthropic is just comically far ahead right now.

On the Deep SWE coding benchmark, Sol on its lowest setting scored the same as Opus 5.5 at maximum effort: 21 cents versus $14.65, about 73 times cheaper. Theo says the benchmark is far from a perfect measure of day-to-day coding, but the gap is worth noticing.

This model on low costs 21 cents versus Astra on low costing $1.46 and Opus 5.5 on max getting the same score for $14.65. While I will gladly admit that Deep SWE is far from a perfect measure of how good a model is at day-to-day code work, the fact that 6.1 Soul is scoring the same as Opus and is also 73x cheaper is

Skills in this one.

Opus builds, Sol reviews

What it is
Use the two models together. Opus 5.5 writes the code. GPT-6.1 Sol reviews its work and investigates the codebase. Theo already has Sol callable from inside Claude Code, Anthropic's coding agent.
Why Theo cares
He finds Sol way better at reviewing and digging into the details, and Opus a more pleasant collaborator that is much better at implementing code without getting blocked constantly.
How to try it
  1. After your agent finishes a change, have Sol review it before you merge.
  2. Ask your agent to set up Claude Code so Opus can call Sol to review its work and investigate the codebase.
the details now. I like using them in tandem because I find Soul to be way better at reviewing and digging into the details, but I find Opus a more pleasant collaborator and significantly better at actually implementing code without getting blocked constantly throughout
config because I already have it set up so that I can use Soul inside of Claude code because I want to make sure Opus knows this is the model to have review its work and investigate the things going on in the code base. This is a

3D models in Blender from a screenshot

What it is
Give Sol a screenshot of something you want modeled in 3D. Tell it that it has Blender, the free 3D modeling app, available from the command line, and ask it to build the model.
Why Theo cares
Sol has no design taste, but Theo says it does a 'pretty damn good job' at 3D modeling this way.
How to try it
  1. Pick a screenshot of an object you want in 3D.
  2. Tell Sol it has Blender over the command line and ask it to make the object.
admittedly really good at Blender. If you give it like a screenshot of a thing you want it to model in 3D and say, "Hey, you have Blender over the CLI. Go make this." It will, and it'll do a pretty damn good job.

Advice

If your work fits what Sol does well, Theo says you should probably use it for everything. If it doesn't, stay on Opus and give Opus the ability to call Sol for the tasks where Sol is better.

If your work fits within the things 6.1 Sonnet does well, you should probably use it for everything. But, if your work doesn't fit in it particularly well, you should probably keep using Opus and maybe give Opus the ability to call 6.1 Sonnet when it should for various tasks.

Hot takes

Sol is 'unacceptably garbage' at UI and front-end design and has no taste. If you want front ends that don't suck, spend your money elsewhere.

His reasoning: His test game looked beautiful, but Sol surrounded it with 20-plus bits of unnecessary text, and it felt awful to play. He calls it a regression he can't believe OpenAI hasn't fixed.

fixed this problem. This model is unacceptably garbage at UI. It has regressed again. And if you're looking for a model that can make front ends that don't suck, go spend your money somewhere else because it should not be spent here. This model sucks at front end. It sucks at design. It has no

OpenRouter's new automatic router, which picks a model for each request to save money, is 'an incredibly stupid idea'.

His reasoning: A model that can't reason, given a prompt with no context, can't judge how hard a task is. In his runs it sent about 60% of requests to DeepSeek V4.1 Flash, a small cheap model.

idea. Once I started running it against benchmarks, I confirmed it's an incredibly stupid idea. It turns out a model that cannot reason, that is given a prompt into no context, cannot make a good decision around how hard the problem is. And Jev Router ended up

This release may mark the start of the end of the era when AI subscriptions are heavily subsidized.

His reasoning: Sol's low prices almost certainly cut into OpenAI's margins. If OpenAI kept subsidizing subscriptions at the same level, the electricity behind your subscription would cost more than you pay, so he gets why they have to change it.

here. That also means that if they were to keep subsidizing the same level that they were on the subscriptions, that your electricity cost for your sub would be more than you're paying. So, I get why they have to change this. They'veawesome release and an unbelievable price for what you're getting, but this does potentially mark the start of the end of the subsidization era. So, make

How Theo thought about this

To get an unbiased read, Theo had Opus review Sol's work under a different name, so it didn't know which model it was judging. Opus quickly rated it a frontier-tier model, and Theo used that blind judgment alongside his own tests.

particularly fun because I had Opus 5.5 review this model with a different name, obviously, so it didn't know what it was. I went and edited the history after. And it concluded very quickly this was a frontier tier model. Its

He uses models to check each other's blind spots. Opus finished the Rust port, but when Sol audited it, Sol found that 1.3 million of the 1.8 million lines were unused. Opus had rewritten the older agents' code from scratch and left the old code in place without noticing.

what it brought up was that of the 1.8 million lines of code, 1.3 million were not being used. The reason was because Opus concluded whole of the code from all the other agents was useless slop that had no chance of being recovered, and it chose to rewrite it from scratch itself in another crate. So, on one

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.