Token Gremlin

Tracking the AI frontier: models, agents, research, leaks & pre-release signals. I test, investigate and build.

[Token Gremlin @TokenGremlin]

@newborec I'm sorry, I didn't mean to make anyone uncomfortable. Thanks for the criticism. 🫡

[Token Gremlin @TokenGremlin]

Thanks for the reminder, Tibo. ChatGPT looks so much better now.

[Tibo @thsottiaux]

Reminder you can customize the appearance of chatgpt on web and make it soothing

https://t.co/an6R8iY5Ix

[Token Gremlin @TokenGremlin]

@newborec Bro, it's just a joke, don't worry. ♥️

[Token Gremlin @TokenGremlin]

Day 21:

A reset, and the biggest confirmation yet that we'll be getting a new OpenAI model soon, within Tibo's 28 Days window.

And that model is (almost certainly) GPT-6.1 Terra (unless they change the name), cheaper than GPT-6.1 Sol and more powerful too, competing directly with Opus 5.5 in raw performance, but beating it on price!

[Tibo @thsottiaux]

Day 21/ Wonder what happens the remaining seven days. But today is a reset.

[Token Gremlin @TokenGremlin]

@AldenAleera Not so far. Tell me more about it, and I'll try to find out if others are having similar experiences.

[Token Gremlin @TokenGremlin]

Are you thinking about canceling your ChatGPT subscription?

[Token Gremlin @TokenGremlin]

@Curline1222 Roon.

[Token Gremlin @TokenGremlin]

Good news from OpenAI: reverse engineering your own TV with Codex isn't against their rules. The blocks people hit were accidental and are getting fixed.

So go ahead. Customize your launcher, kill the ads, and make it faster with any ChatGPT model.

Claude is still a gray area though. Try it there and it'll assume you're a Chinese state hacker breaching a power grid.

[Token Gremlin @TokenGremlin]

@vishalsingh2972 You got lucky. Some of my colleagues are already running into the safety restrictions. The system routes their requests to a weaker model, and in some cases, even flags them as attempted cybercrime.

[Token Gremlin @TokenGremlin]

Take a few days to explain AI to your family. What it is, what it's for, and how it's already changing the world. Use simple examples and show them live. Sometimes the best way to improve your life and theirs is to not be lazy about teaching. They're not in the same X bubble you are.

[Token Gremlin @TokenGremlin]

A lot of AI companies are about to disappear, the AI debt bubble is about to burst, and people in my everyday life (A LOT OF PEOPLE!) don't even know what Opus 5.5 is. It genuinely feels like I'm living in two completely different worlds at the same time.

[Token Gremlin @TokenGremlin]

A lot of AI companies are about to disappear, the AI debt bubble is about to burst, and people in my everyday life (A LOT OF PEOPLE!) don't even know what Opus 5.5 is. It genuinely feels like I'm living in two completely different worlds at the same time.

[unusual_whales @unusual_whales]

BREAKING: Elon Musk has said:

"As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers. That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.”

[Token Gremlin @TokenGremlin]

I honestly feel like Anthropic has won. This year belongs to them, and they've pretty much taken over the market.

And even if OpenAI fights back with GPT-6.1 Terra, Anthropic already has Fable 5.5 ready to go.

To be honest, if Terra doesn't beat Opus 5.5 SOON (!!!!), no enterprise discounts or usage resets are going to save OpenAI from spending months in second place.

[Andrew Curran @AndrewCurran_]

I don't take money from any lab. I'm saying this impartially; the number of people on my feed over the last two weeks saying they are switching over to Anth because of Opus 5.5 is like nothing I've seen before. And even if someone does beat Opus, they have Fable 5.5 ready to go.

[Token Gremlin @TokenGremlin]

New details about GPT-6.1 Terra:

- Cheaper than GPT-6.1 Sol, which already costs $2 / $10 per million tokens
- More powerful than GPT-6.1 Sol
- Already spotted in Microsoft's eval configs, where 18 test setups switched from GPT-5.6 Terra to gpt-6.1-terra
- Release could land in the next few days

Yes, you read that right. A model one tier below Sol, beating Sol, for less money.

That breaks OpenAI's own naming logic. Luna, Terra, Sol, Astra used to mean bigger and smarter as you go up. Now the names look more like price points than power levels.

If this holds up in real-world testing, Terra could be a much bigger deal than its name suggests.

OpenAI might actually be cooking here.

[Token Gremlin @TokenGremlin]

GPT-6.1 Terra is already showing up in publicly visible backend tests, and OpenAI is preparing for its release.

But honestly, I'm struggling to see the point of launching it right now.

The model hierarchy is Luna, Terra, Sol, and Astra. Terra sits below Sol, so it wouldn't make much sense to expect it to outperform GPT-6.1 Sol.

And with everyone waiting for OpenAI's answer to Opus 5.5, releasing a lower-tier model feels like a strange move.

Maybe Terra has something special to offer, but right now, this isn't the kind of release I've been hoping for.

[Token Gremlin @TokenGremlin]

Hey @grok remove the worst company from the image.

[Token Gremlin @TokenGremlin]

@InfiniteHexx I honestly wish that were true. But they're still using the "gpt-6.1-terra" name.

I'm just hoping this is all some elaborate misdirection and they switch it to GPT-6.2 or something at the last minute. LOL

[Token Gremlin @TokenGremlin]

GPT-6.1 Terra is already showing up in publicly visible backend tests, and OpenAI is preparing for its release.

But honestly, I'm struggling to see the point of launching it right now.

The model hierarchy is Luna, Terra, Sol, and Astra. Terra sits below Sol, so it wouldn't make much sense to expect it to outperform GPT-6.1 Sol.

And with everyone waiting for OpenAI's answer to Opus 5.5, releasing a lower-tier model feels like a strange move.

Maybe Terra has something special to offer, but right now, this isn't the kind of release I've been hoping for.

[Token Gremlin @TokenGremlin]

@adityaa_behera Exactly.

[Token Gremlin @TokenGremlin]

This isn't really the future. The pieces are already on the table.

Opus 5.5 can reason through almost anything you throw at it. Dessa can sound more human than most humans on a phone call. Caller ID spoofing is decades old.

Nobody has to invent anything new. Someone just has to wire it together into an agent that never sleeps and calls thousands of people a day.

[Nikita Bier @nikitabier]

The perfect storm is frontier models finding obscure security vulnerabilities at breakneck pace while voice models become indecipherable from human speech.

When a model hits a roadblock, it will just spoof Caller ID as your bank and call you, pretending to be your banker.

[Token Gremlin @TokenGremlin]

@CodeByNZ

[Token Gremlin @TokenGremlin]

If OpenAI had already released GPT-6.1 Astra, we'd probably have Fable 5.5 by now too.

Anthropic only moves fast when OpenAI puts pressure on them, and that's exactly why I'm rooting for a new OpenAI model next week.

[NZ ☄️ @CodeByNZ]

@TokenGremlin rooting for openai so i can get a new claude model is a strange way to live but here we are lol