🚨codex users!!!! I tested GPT-6 Sol on $20 plan I set the effort to "MAX" I am NOT happy with the results - 15 Million tokens - 7% weekly limit used - $4.21 API cost - 35 minutes of work So... you are getting 857 Million tokens of GPT-6... Sol And wierd part is... only $250 API usage / month?? Claude was $1100 / month with Opus 5 at Max at $20 sub AND Again!! IT IS EXTREMELY SLOW!!! it did NOT complete the task!! I specifically mentioned it to test in THIS SPECIFIC way it only tested "builds" and called it a day Tbh, this is my 4th test with Codex... and I'd prefer Claude (with Opus 5) NOTE: This was a test, I don't use gpt-6 or Opus 5.5 on max for my own workflows... this just shows how much you are getting in "tokens" for the "price" you are paying.show more

This is the config for anyone who wanna dive deep

I understand setting Sol to max effort but don't set Opus 5.5 to max effort. It actually gives better results at medium effort than any other reasoning. Use Opus 5.5 on medium on your $20 plan and you won't need to buy Claude Max plan at all.

Yeah I know that but I set these efforts to the Max plan so that I can test out the maximum usage that you can get on each $20 plan. That's all.

Absolutely. I think that Claude's usage might be a bit higher than Codex's -- that's what you found out too.

You put it on token burner mode and complain that it burns tokens Try anything less than max Not saying it’s good or anything but you can’t complain that it burns tokens on token burner mode

It's only for tests bro... I am only calculating the amount of work you get for the price you pay

Alright my mistake I interpreted it as complaining

Guys... I am gonna test out every provider's limits now to make it MORE clear It takes effort so... Thanks!

cheap sol. still burns the week

Yeah man, I am fed up with all this

All it took was 1 prompt using GPT6 Sol on medium.

Sorry bro 😐 Switch to Claude maybe

yeah it's time to move back to claude,i miss the old opus 4.6 and i think 5.5 has the same feel

It DOES have the same feel bro

Wow curious how you arrived at 1100$ with Opus 5.5?

I didn't test Opus 5.5... only Opus 5. Its also posted on my account... you'll love to see it

they are not only shit about there usage but there support team is full of AI, even being 20x pro user i regret having it. despite 20x plan is in paused

OpenAI is making so many worse decisions ngl

month ago i was huge huge fan of @thsottiaux and @sama every single user response is genuinely heard. right now they even scam the banked reset. we are getting 30/40% less usage in banked reset

@thsottiaux @sama Exactly bro... I was a fan too but now... meh

I know!! astra cooked but sol just like yea imma head out and say i cant do it

Sol came later so at least it should have been better than Astra but nah...

I guess Sammy wants to address the astra huge consumption complaint, we Plus users can stay in the loop. I guess any user count at this stage. Astra is amazing though

Yeah it’s exactly the same usage as the old one, it’s rubbish

Honestly... I can't see how I will use Codex in future

yep, how crazy that Anthropic has a much better value for $20 now 😭

I know man... I was a fan of codex just 1 month ago... now its disappointing

My Claude Max 20x already works out to around $1,500/week in API-equivalent usage. And you’re telling me Pro gets $1,100/month? Doesn’t that make it pretty obvious that something is wrong with the calculation?

20x plan gets around $8k worth of usage... And it's not 20x the Pro plan that is why Anthropic got sued

In practice, Claude Max 20x works out to roughly $6,000/month in API-equivalent usage. Based on earlier testing, Max 5x is about half of that, so roughly $3,000/month. And Pro is about 1/5 of Max 5x, which would put it closer to $600/month. So I honestly don’t understand how you ended up measuring $1,100/month for Pro.

Here is the test... Go check it out... I tested with proof how we are getting $1100 of usage monthly

very bad 😭 but opus 5.5 is extremely slow on max

No way!! I'll have to test it out too

I agree. I am a plus user and single prompt entire 5hr quota over in 25 min and work incomplete. I used GPT-6 SOL Medium

I used GPT-6 SOL on max It usage is "okay-ish" if you don't code

I’ve been running sol on high and xhigh and I had a completely different experience. Feels like 10x usage compared to astra, I ran out after ~2 hours

I am not saying that it's a bad experience. Yes I can agree with you that GPT-6 Sol can run for 2 hours straight with high effort but the problem is that it is slow. The amount of work that you got done in 2 hours is nowhere near the amount of work that you would have gotten done with Claude Opus 5. Also the amount of usage that you get on Claude and Codex is completely different. Claude gives you a lot more usage. 32 Million tokens vs ~18 Million for same 10% weekly is crazy difference!

At least we get more resets for codex… you can’t have everything in this ai world🥲

it is very slow and anyway even without using astra my usage is already 60% wtf (im only using luna mostly and sol medium / high)

Bro, that is so rough. You might be having some kind of glitch.

Maybe I don't know, bro, but what's happening with Codex? I'm on Windows. To get these models, I even had to clear my cache manually. It means that even when I update an app, I don't get the models they have. So many issues that haven't been solved.

Yeah... The Desktop app is slow af... That's why I used CLI ... Might as well use CLI cause it's also a little bit better at caching... Personally speaking

OK, bro. Thanks, I've got it! I'll give it a try. I'm using the app only because I like reading the text from it, but I'll give the CLI a try again!

No problem... lemme know your experience

I’d think performing the exact same task/prompt/context etc on these 2 specified models at max would give us a better picture.

I will test it out for you... No problem... Next post is exactly that

OpenAI might have a compute problem. With Opus 5.5 being this capable they can't cutting limits like this without loosing subscribers. Thanks for the test. Real world use cases are the only benchmark we can trust right now.

No probs bro... I will test out Claude Opus 5.5 too

Tokens are fuel, not miles: a cheap, high-throughput model can still cost more in time if it doesn't finish the task.

That is what I am saying 😂

remember there is batch request in sol which is good for only bulk changes or scanning giant codebases.

Yeah but it doesn't change the fact that we are getting 1/2 usage fr same price

Yes, API prices seem more reasonable sometimes

but rn now they are not 😢

Just last week it was 350$ of usage. You mean its 250$ now? 😐

Yeah 😂 Move to Claude!!!

So the result is use Claude 🔥 it’s always the best

Yeah... I think we should switch to Claude for now Don't know if this will be the same for next month

I think Claude is always better in every case

It wasn't until August... I think they did something like this: 🤣 OpenAI did all the bad things themselves

Yea gpt 6 is fricken slow. I don't see any improvements really. Opus all the way.

It's only an update, rather than an upgrade 😂

Exactly. Well they said devday is huge and that's coming Tuesday next week. Time will tell.

Really man, I think that'll be flop too

Could you please check the performance of Codex 20$ 6Sol and Claude 20$ Opus 5.5?

Claude Opus 5.5 is on the way... I a travelling right now... Will test it when I get back.

last time i tested claude fable 5 vs gpt 5.6, gpt was way better.

Hahahahaha... GPT 5.6 Sol can't touch Fable 5 ngl... I don't know how you tested it... Lemme know about that

yikes, that's rough. speed and accuracy are everything.

Yeah man... I was a codex fan... now, I am not

Fair. But remember, chat + images also count ontop of that. Codex / Work limits are apart from Chat + Image limits. Claude has a total limit, not unified. 5.6 Sol on high - no matter how many tokens it outputs. Perfect to check your codebase in a zip... 🫢

The thing is simple. I don't use chat or images... Simply give me full usage of what I want... But even that won't be enough ngl

damnn thats sounds like a steal 🥀 857 million tokens for $250 and it still didn't finish the task in 35 minutes

Not a steal... but a scam

The denominator is doing all the work here. 35 minutes for 7% implies roughly 8.3 hours per week if the burn stays linear; 857M tokens is extrapolated throughput, not an allowance the plan guarantees.

💯💯💯

yeah mate, disappointed as well. Have been using it as my daily driver today at High and only about 5h of use later it's down more than 30%. I'm still on the $100 codex plan with one reset left but honestly as much as I dislike anthropic I might switch back after this month

Feeling sorry for that bro... Might as well try Claude... I think it's better with $20 plan... Can't say anything for $100 plan

focus 100% on cost per task not on tokens, and stop using max

I only tested sir... The cost per task is 2x the price of Claude if we look at the $20 we pay them (not API usage)

Max effort is not free intelligence. Token burn without an outcome is just an expensive thinking face.

Opus 5 always gave 2x more usage than sol 5.6 in every plan. Openai made up for it with the resets, I realized how bad they were when they gave a reset after 10 days and I had to use 200% of my weekly quota with a banked reset

They are making themselves worse like this 😐 I hope they'll learn