Really excited about AGI, learning more every day about LLMs and the future of technology, and sharing my own thoughts along the way
- Tweets 2,177
- Following 1,088
- Followers 2,719
- Likes 238,400
I burned through the entire weekly 20x plan in a single day, and I got around 1.3B tokens with GPT-6 Astra, compared to roughly 3.5B tokens per week with GPT-5.6 Sol, Extrapolating from that, the $20 plan should get around 70M Astra tokens per week, while the $100 Pro plan should get around 350M
But as contradictory as it sounds, you can actually get a lot more tasks done with Astra than with Sol despite that huge difference in raw tokens, I explain exactly why, with all the numbers, in my article here!
As Greg Brockman, OpenAI’s co founder and president, put it price per token no longer matters; what matters is the price per completed task!
12
4
56
20,315
Looks like GPT-6.1 Sol usage got nerfed pretty hard, Up until Friday I could run around 10 projects in Ultra and it would only eat about 3% of my limit per hour, but now doing basically the same thing burned through my entire x20 plan in less than 12 hours :( Before it genuinely felt almost unlimited, and now the limits feel way tighter, Luckily I still have another x20 account, so for now I’m mostly sticking to Very High there, but even then I’m noticing a pretty obvious drop in how much usage I’m getting. I thought the crazy limits would last a little longer lol. At least as a small consolation, I’ve got my Dot working with my GitHub 24/7 now
17
3
68
5,973
🚨 Gemini 4 Argon is probably coming this week, Sonnet 5.5 and Opus 5.5 are already showing up in Antigravity, and I tested Opus 5.5 Medium myself on the $20 plan for about 20 minutes. It ended up using around 50% of my weekly limit, and it also completely burned through the separate 5 hour limit for Claude and GPT models, Maybe Opus 5.5 Slow or Sonnet 5.5 Medium will give us a lot more room to use them, but honestly, everything happening right now makes it feel like Gemini 4 Argon is extremely close. If it really drops this week, Google could be about to bring out another serious frontier model and jump right back into the race
17
6
246
26,728
Codex limits reset 3 hours later, on one of my accounts, I managed to burn through all 62,500 credits in a single day with 10 projects running in Ultra, multiple agents working at the same time, and GPT 6.1 Sol handling most of it without errors
So tonight, another wave of Ultra runs is coming 😅
Huge thanks to OpenAI, This is honestly the first time I’ve been able to use Ultra mode without constantly worrying about hitting my limits
5
1
30
6,042
🚨 Fable 5.5 XHigh vs GPT 6.1 Sol Ultra
Fable 5.5 now seems to be routing through both the web app and Claude Code for most of us, On my Claude Pro x5 plan, this test used around 7% of my weekly limit without any agents, Running the same kind of test on Codex x20 used only about 1% of my weekly limit
Quality wise, both were really strong. In terms of speed, they were also pretty close, with both taking roughly 30 minutes to finish, I’m running a lot more tests in parallel right now to see whether there’s actually a meaningful difference between them once you push both models harder
30
15
388
70,062
Here I leave an even more exhaustive and complex test, really impressed!
Fable 5.5 XHigh vs GPT-6.1 Sol Ultra , the quality from both is honestly incredible, I don’t think any other companies besides Anthropic and OpenAI are able to make results look this realistic
Speed wise, Fable finished almost twice as fast, but it used around 5% of my weekly limit on the Pro x5 plan Codex, on the other hand, only used about 1% on the x20 plan , In terms of pure quality, I think they’re extremely close. Personally, though, I liked Fable 5.5 a little more, especially the hyper realistic zoom, That part looked incredibly good
Right now, it honestly feels like Anthropic and OpenAI are at least 6 months ahead of the rest when it comes to this kind of quality, I’m still amazed by how good these models have become, especially considering how low the cost is
2
486
Fable 5.5 XHigh vs GPT-6.1 Sol Ultra , the quality from both is honestly incredible, I don’t think any other companies besides Anthropic and OpenAI are able to make results look this realistic
Speed wise, Fable finished almost twice as fast, but it used around 5% of my weekly limit on the Pro x5 plan Codex, on the other hand, only used about 1% on the x20 plan , In terms of pure quality, I think they’re extremely close. Personally, though, I liked Fable 5.5 a little more, especially the hyper realistic zoom, That part looked incredibly good
Right now, it honestly feels like Anthropic and OpenAI are at least 6 months ahead of the rest when it comes to this kind of quality, I’m still amazed by how good these models have become, especially considering how low the cost is
17
9
199
20,157
🥳 We’re getting a reset for all ChatGPT users tomorrow! , On top of that, GPT 6.1 Sol, OpenAI’s most efficient model yet, is now running much faster than before after they added more compute to handle the demand, It’s honestly great to see the extra capacity kicking in
Here are the exact reset times for each country:
🇺🇸 Estados Unidos 10:00 AM PST
🇮🇳 India 11:30 PM
🇩🇪 Alemania 8:00 PM
🇧🇷 Brasil 3:00 PM
🇯🇵 Japón 3:00 AM
🇬🇧 Reino Unido 7:00 PM
🇰🇷 Corea del Sur 3:00 AM
🇮🇩 Indonesia 1:00 AM
🇫🇷 Francia 8:00 PM
🇪🇸 España 8:00 PM
I think it’s much better when they announce these resets ahead of time. It gives us a lot more time to actually use our tokens properly, especially since we still have more than 14 hours left!
Global reset landing tomorrow 10am PST for all paid ChatGPT accounts. Apologies for the slow start with GPT-6.1 Sol, it's now back to running at expected speeds after the massive load spike in the first two days.
1
1
15
2,305
🥳 Grok 4.7 is now available in Chat mode!
One of the coolest things is that you can actually have it interact with your Grok Bots to improve them, optimize them, and keep refining how they work
I’ve been testing it in Heavy mode, and so far it feels noticeably faster than Grok 4.6
Chat mode also feels almost unlimited, so I’m going to be testing it pretty heavily and using it to optimize all of my Grok Bots, Really liking
3
5
48
4,312
Grok Primary Bot? This just came out, and Grok itself describes it as a “primary assistant” that can also control other bots
It honestly looks really cool, I’ve already been using it to create new bots and improve some of my older ones, and I’m really liking the feature so far
You just need to update the app on your Mac or iPhone and it should automatically show up
Hopefully everyone gets access to it soon, Grok keeps getting better and better
2
1
13
1,115
I tested Gemini 4.0 Pro on Arena when it was available, and now Gemini 4 Argon has finally been announced, What surprised me the most was not even the raw quality, but the way it teaches and explains things, I have not seen another model use learning methods like this, It felt much more interactive, easier to understand, and genuinely more pedagogical, so that is where I would give Gemini a real competitive advantage, It was also extremely fast, Now the big question is whether it can be as efficient as the Sol models. At the end of the day, what really matters is not just how many tokens you get per dollar, but how much it actually costs to complete a task successfully, Efficiency is the key if they really want to compete, Either way, I am really excited to see it become available to the public soon
6
19
393
40,985
Sonnet 5.5 XHigh vs GPT-6.1 Sol Ultra, this was really impressive Sonnet ended up costing about twice as much to run, but it actually produced the better result in this test, My impression from this test and several others is that GPT 6.1 is still insanely efficient. Sonnet is very good, but it tends to think for much longer and burns through usage limits much faster. To put it into perspective, this single run used around 10% of my weekly limits on the $100 plan, while Codex only used about 1% running in Ultra mode, I have also been doing some 3D tests that I will share soon, and the results there were completely unexpected
7
7
82
7,497
62,500 credits were just given to every legacy $200 Codex Pro account, and apparently that can translate into more than 81,000 local messages at the absolute maximum, That is actually insaneeee with GPT 6.1 Sol, the plan genuinely feels unlimited at this point. I am starting to worry that I will not even be able to burn through all these credits and my banked resets in time This is exactly why I love OpenAI so much, They are always transparent about what they are doing and they actually seem to care about their users!
7
2
49
8,719
Your Dots already come with Blender installed on their own machines, so you can basically ask them to work for you 24/7 without using your computer’s memory or eating into your limits. What really surprised me is how fast and smooth everything feels, and how perfectly it syncs with the Codex app, You can call your Dot while it’s working, talk to it, interrupt it, ask for proof of progress, or check exactly what it’s doing, and it genuinely feels like interacting with a real agent that’s always there working for you, This is honestly much closer to the kind of AI agent I’ve always wanted, Really excited about my Dot :)
5
2
54
3,931
GPT-6.1 Sol ULTRA ran for 25 minutes and only used 1% of my weekly quota, It’s seriously fast and powerful, almost like Astra, but much cheaper to run, For me, this is the kind of model you can actually use 24/7 without constantly worrying about limits, From now on, it’s probably going to be my default model, Honestly, out of everything announced at DevDay, this might be my favorite update. In my experience, it feels like the most efficient model OpenAI has released so far
72
40
1,005
140,805
UPDATE: it ran for a full hour and the usage is still exactly the same as what I showed in the video, This is honestly insane, Astra level performance that you can basically run 24/7?
GPT-6.1 Ultra ran for a full hour with agents and only used 1% of my weekly quota on the x20 Pro plan, In some of my early tests, it’s actually performing better than Astra while being way more efficient, My first coding tests have been seriously impressive and really fast, but I’m still running several benchmarks in parallel to see if it can actually compete with Opus 5.5. I honestly didn’t expect we’d get a model this fast, Easily one of the best releases this month
2
1
38
8,264
GPT-6.1 Ultra ran for a full hour with agents and only used 1% of my weekly quota on the x20 Pro plan, In some of my early tests, it’s actually performing better than Astra while being way more efficient, My first coding tests have been seriously impressive and really fast, but I’m still running several benchmarks in parallel to see if it can actually compete with Opus 5.5. I honestly didn’t expect we’d get a model this fast, Easily one of the best releases this month
GPT-6.1 Sol ULTRA ran for 25 minutes and only used 1% of my weekly quota, It’s seriously fast and powerful, almost like Astra, but much cheaper to run, For me, this is the kind of model you can actually use 24/7 without constantly worrying about limits, From now on, it’s probably going to be my default model, Honestly, out of everything announced at DevDay, this might be my favorite update. In my experience, it feels like the most efficient model OpenAI has released so far
8
5
95
26,450
Codex Pro 20x , I just tested this on my second account and got close to 17B tokens of usage between Astra and GPT-6 Sol, basically the same as what I got on my first Pro 20x account, That works out to roughly $500 worth of API priced usage per day on a $200 plan, which will now drop to around $250 per day, I honestly don’t really understand why they announced this today out of nowhere. Wouldn’t it have made more sense to announce it after the DevDay products were released? And apparently the new $500 plan is being announced today too!
The $200 Codex Pro plan now gives you roughly $4,000 worth of API priced usage around 2B tokens with Astra + 6B with GPT-6 Sol, That’s over $16,000 per month at API pricing, and that’s not even counting the resets, According to Tibo, though, the gap between API pricing and subscription plans will get much smaller in the future, So eventually, a $200 plan could give you something much closer to $200 worth of API usage, enjoy this era of heavily subsidized compute while it lasts, I absolutely love Codex, it’s probably my favorite tool to use every day, and I’m really glad they’re being transparent about this and now it’s also confirmed that there won’t be a 5 hour limit, which is amazing!
6
3
63
25,405
And yeah, SemiAnalysis was already estimating around $14,000 worth of API usage for the $200 Codex plan about four months ago, so this isn’t really anything new
What Tibo seems to be saying now is that those limits would be cut roughly in half, which is the part that actually matters x.lingyaoai.com/SemiAnalysis_/status/2…
Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)
5
901
The $200 Codex Pro plan now gives you roughly $4,000 worth of API priced usage around 2B tokens with Astra + 6B with GPT-6 Sol, That’s over $16,000 per month at API pricing, and that’s not even counting the resets, According to Tibo, though, the gap between API pricing and subscription plans will get much smaller in the future, So eventually, a $200 plan could give you something much closer to $200 worth of API usage, enjoy this era of heavily subsidized compute while it lasts, I absolutely love Codex, it’s probably my favorite tool to use every day, and I’m really glad they’re being transparent about this and now it’s also confirmed that there won’t be a 5 hour limit, which is amazing!
Hi,
Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers, but together with it we are also changing how we calculate the usage for it. In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan.
Now that it's said, let me explain why this is happening and why you will still get more work done than if you were on the Pro $200 subscription one month ago.
(a) We didn't want to compromise in other ways and are committing to not reintroducing the 5h limit, so that you can fully use the weekly usage when you want.
(b) On the subscription, we guarantee that over time you always get more work done and with an increasing level of quality. This means that you will continue to get more value per dollar spent as a result of models getting more efficient and us passing down the improvements in the form of API price reductions.
(c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc). Instead we want to continue to both rapidly reduce prices and increase capabilities of models on the API. This week we introduced GPT-6 Sol and GPT-6 Luna at 50% of their previous price. Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent.
(d) Tomorrow, we are adding more things to the subscription that won't draw on the usage, I won't reveal what that is yet.
I wanted to be transparent before all the big announcements tomorrow. Lots of new exciting things are coming to the subscriptions that will make it super compelling, but I wanted to make sure to share this change ahead of time so you can all understand it before we shower you with good news.
Codexingly,
Tibo
86
15
330
95,045
This is the usage report from my second account using OpenUsage, and the numbers are really close, They basically match the limits I’m seeing on my first account too,I use the first one on my PC and the second one on my Mac
Codex Pro 20x , I just tested this on my second account and got close to 17B tokens of usage between Astra and GPT-6 Sol, basically the same as what I got on my first Pro 20x account, That works out to roughly $500 worth of API priced usage per day on a $200 plan, which will now drop to around $250 per day, I honestly don’t really understand why they announced this today out of nowhere. Wouldn’t it have made more sense to announce it after the DevDay products were released? And apparently the new $500 plan is being announced today too!
8
5,552