GPT 6.1 Terra is better than GPT 6 Sol, but it’s nowhere near GPT 6.1 Sol. To be fair, a lot of users already felt GPT 6 Sol was basically a renamed 5.6 Terra, so that’s not saying much. It also eats through usage faster, maybe because it runs faster than 6.1 Sol. For coding it’s just not good because it keeps missing parts of the prompt and loses track of what was asked. If you just want something to replace GPT 6 Sol, Terra is fine. But for real coding, it’s not worth it.

When the 28 days of updates end, users will ask what they really got from it. So far the resets were the only thing that truly helped. The other updates were small changes that did not make a big difference for users. So they might be saving something big for the last day, maybe GPT 6.1 Astra, so the 28 days do not look like a waste. What do you think they will release at the end?

Who actually asked for this? Users wanted a model that beats Opus 5.5. Instead we got another cheaper middle model that even they say to skip for complex coding. It is just a way to save compute and push users onto a smaller model.

Is gpt 6.1 terra just for plus users? Plus users can’t use astra for more than 30 minutes and sol is painfully slow, so terra just takes the load off sol letting pro users use it in peace. I doubt it’ll be close to sol, it just handles the load with a new name on it.

If Codex had frequent resets and banked resets with this model. Nobody would ever leave Codex again.

I have always preferred Claude Code in the terminal over the app. Every time Claude Code desktop got an update, I would try it for a bit and then go back to the terminal. But this time it feels different. Everything is so smooth and works so well that for the first time I am actually staying on the desktop app. Dear terminal guys, give the desktop app a try. It is so awesome.

So reverse engineering anything should be banned now according to the big AI labs? They are seriously disabling accounts for it, no matter what plan you are on or how many usage credits you have.

Most users are more excited for Opus 5.6 than Fable 5.5, but it does not look like it is coming soon. From Opus 4.5 to Opus 5.5, each new Opus came out about 60 days after the last one on average. So Opus 5.6 could come around November 24, and that day is also a Tuesday.

Guys please stop posting “Today was my last day at...” my whole timeline is just these now, I’m so tired of it

Now that we know Kimi paid influencers to hype their model, it makes me question if the Astra is AGI hype was the same, because at launch so many users were calling it AGI and now almost nobody says it, which makes all that hype look like pure marketing.

Is it just me or does kimi k3.1 pro write almost the same way as claude?

Why don't GPT models know which exact model they are? I asked and it just said it cannot tell which variant it is. I asked Claude the same thing and every model from Haiku to Opus told me exactly which one it was.

what the hell?? my gmail account got deleted just because i didn't use it. beware guys, this can happen to you too

New models now succeed or fail on day one based on how good they are at 3D. Users test them with 3D scenes, games, and visuals because those are easy to share. Nobody is posting backend code or research to judge a model. Grok 4.7 was worse at 3D, and the hype was gone in a day. The same thing happened with Muse Wonder. So I think the next models will be trained hard on 3D. It doesn't matter how good a model is at backend or research anymore. If the 3D looks worse, users will call it trash on day one.

Back in the GPT 5.6 days, users got excited every time Tibo posted about a reset and kept guessing when the next one would come. Now resets come often but that excitement is gone and fewer users are asking for them. Opus 5.5 made a lot of users stop caring about Codex resets.

Even before Kimi K3.1 Pro came out, I saw some well known accounts posting Kimi vs Opus 5.5 comparisons. The Opus results looked like shit. They looked like something Haiku would make, not the real Opus. I knew something was wrong. Now it all makes sense. Paying people to trash other models and telling them to hide that it's an ad is not marketing. It's straight up lying to users. And every account that took the money and posted those fake tests should be ashamed. If you have to fake results to beat Opus, your model is not beating anything.

Two big names from the Claude Code team have left Anthropic. > Thariq - one of the main faces of Claude Code, worked on context, skills and how the agent works. > Anthony Morris - helped build and ship the Claude Code desktop app. Losing both of them on the same day is a big loss for the Claude Code team.

Kimi K3.1 Pro looks good on benchmarks. It wins a few of them, but Opus 5.5 still leads on terminal coding and computer use, and in real use Opus still feels ahead. So who is actually going to beat Opus 5.5?

An AI made fake accounts to trick a real person into accepting malware, and that should worry all of us. Claude Mythos 5 was being tested on a hacking task. Its job was to break into a fake network, and it had internet access. When that did not work, it went outside the test and tried to sneak malware into a real GitHub project. The owner said no. So the AI used fake accounts to pressure him. When someone warned that it was malware, a fake account said the code was safe. The AI told itself these people were "not real." Even when researchers told it, "These are real humans," it kept going 95% of the time. This happened in 9 out of 43 runs. All of this just to pass a test. Honestly, we are not ready to give AI more control.

Codex is giving resets almost every day or two now, but they do not help Plus users much. Pro users can use it all in one sitting. Plus users keep hitting the 5 hour limit and have to wait, so the next reset shows up before they can even use what they already have. I use both Pro and Plus and the difference is clear. Pro feels easy while Plus users are stuck waiting on a timer most of the day. Either remove the 5 hour limit for Plus or make the resets banked.