AI Thinkers

🚨 I am your unfair advantage...

[AI Thinkers @The_AIThinkers]

🚨 Unpopular Opinion

Benchmarks are just useless. Instead of showing benchmarks solve real life problems with the least amount of hallucinations

[AI Thinkers @The_AIThinkers]

Caude Mythos or Oceanus, whatever they call it, is genuinely insane.

It built a custom meshing engine using Three.js from jsDelivr, all inside an HTML file with low‑effort thinking.

Credit to @Lentils80 and z..AI. Most people don't realize how good this actually is yet.

[AI Thinkers @The_AIThinkers]

We're crossing a line. Human-in-the-loop is slowly becoming optional.

Anthropic ran an experiment with Mythos. In selected research moments, the model chose a smarter next step than the human researcher 64% of the time.

that's the end for mandatory human needs in certain loops.

[AI Thinkers @The_AIThinkers]

They ran it on their GB200 Abilene cluster, which had about 112,000 GPUs since December or January.

So the base model was only a 16‑day training run, plus maybe two months of post‑training.

Now think of 90‑day run with 300,000 GB200 GPUs. That's a different league entirely.

[AI Thinkers @The_AIThinkers]

I just realized how small the training run for GPT‑5.5 actually was compared to what's out there now.

I think: GPT‑5.5 is around 5 trillion tokens, trained at NVFP4 precision similar to what Nvidia used for Nemotron 3 Ultra.

[AI Thinkers @The_AIThinkers]

@OfficialLoganK Lets Goooooo 🚀

[AI Thinkers @The_AIThinkers]

ChatGPT's new memory system is out today for Plus and Pro users in the US. And it comes with twice as much memory.

That means the model can remember more across your conversations.

[OpenAI @OpenAI]

We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time.

Today, that work is rolling out as a more capable memory system in ChatGPT. https://t.co/0MyFKCe2Mu

[AI Thinkers @The_AIThinkers]

in about six months, Anthropic went from Opus 4.5 to Mythos.

and now they're shipping roughly 3x more code per person.

that's a real productivity leap.

[Anthropic @AnthropicAI]

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.

It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx

[AI Thinkers @The_AIThinkers]

@AnthropicAI Let's goooooo 🚀

[AI Thinkers @The_AIThinkers]

BREAKING 🚨 Google says they're building the world's best vibe coding app for Android and iOS.

I'll only believe it when I see it. They've said things like this before.

[Logan Kilpatrick @OfficialLoganK]

we are cooking the worlds best vibe coding app on Android and iOS, it’s going to be so cool :)

[AI Thinkers @The_AIThinkers]

i'm not getting any signal that GPT‑5.6 drops today. Opus 4.8 is roughly on par with GPT‑5.5 overall.

so OpenAI has no real pressure to rush. Unless 5.6 is mostly an efficiency play to cut compute costs.

that's the only reason I'd see them moving fast.

[AI Thinkers @The_AIThinkers]

Let's be honest:

"AI safety" is just gatekeeping, control, and censorship with a nicer name.

[AI Thinkers @The_AIThinkers]

Google just dropped a new Gemma 4 variant. 12B unified, multimodal, and built to run on a laptop with 16GB VRAM or unified memory. Fully open source under Apache 2.0.

pattern I'm noticing: proprietary models which are fast, and open models which are small but seriously capable.

[Google @Google]

Today we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop.

It delivers performance nearing our larger Gemma models with a much smaller total memory footprint, while being small enough to run locally with just 16GB of VRAM. It’s open and accessible for everyone to use under a permissive Apache 2.0 license.

This is all made possible by our new, unified architecture that removes separate multimodal encoders. Here’s how we did it 🧵

[AI Thinkers @The_AIThinkers]

Ideogram just dropped 4.0, and it's the open image-gen model.

It sits at 8th on LM Arena and 5th on Design Arena for text-to-image. Getting close to Nano Banana Pro.

It has accurate text rendering, native 2K resolution, active background transparency, and proper layout control.

[AI Thinkers @The_AIThinkers]

opus 4.8 isn't a bad coding model at all. On ProgramBench, it gets 71.88% raw pass rate with 31 almost-resolved tasks.

GPT-5.5 is at 70.77% with 19 almost-resolved.

but Opus 4.8 costs $36.10 per task. GPT-5.5? $6.95.

that's more than 5x the price for a tiny gain in pass rate.

[AI Thinkers @The_AIThinkers]

The next couple weeks are going to be packed.

Opus 4.8 dropped last week. In the same blog post, Anthropic said a "mythos‑class" model is coming in a few weeks.

And if OpenAI follows their usual release style, GPT‑5.6 probably lands next week.

Going to be fun to watch.

[AI Thinkers @The_AIThinkers]

BREAKING 🚨 Perplexity Personal Computer just dropped for Max and Enterprise Max users on Windows.

If you're on one of those plans, you can finally try it.

Waitlist below if you want in 👀

[Perplexity @perplexity_ai]

Join the waitlist for Personal Computer on Windows:

https://t.co/5SEJhUs59d

[AI Thinkers @The_AIThinkers]

The next couple weeks are going to be packed.

Opus 4.8 dropped last week. In the same blog post, Anthropic said a "mythos‑class" model is coming in a few weeks.

And if OpenAI follows their usual release style, GPT‑5.6 probably lands next week.

Going to be fun to watch.

[AI Thinkers @The_AIThinkers]

from what I can tell, Mythos isn't just "the new Opus." It's a whole tier above Opus.

that also means it'll cost more. So Anthropic still needed to ship Opus 4.8 as a genuine upgrade over 4.7.

but if 4.8 is better than 4.7 but worse than 4.6? Then what's the point.

[AI Thinkers @The_AIThinkers]

ChatGPT just hit 1 billion monthly active users.

Gemini is close at 900M+.
Claude? 56 million.

now guess which one is actually making the most money right now.

makes you think about who's really paying vs. just browsing.

[AI Thinkers @The_AIThinkers]

BREAKING 🚨 OpenAI just dropped "Sites" inside Codex.

You can now generate and deploy a full website without leaving the chat. No tab switching. No deploy buttons. Just build and go.

[AI Thinkers @The_AIThinkers]

from what I can tell, Mythos isn't just "the new Opus." It's a whole tier above Opus.

that also means it'll cost more. So Anthropic still needed to ship Opus 4.8 as a genuine upgrade over 4.7.

but if 4.8 is better than 4.7 but worse than 4.6? Then what's the point.

[AI Thinkers @The_AIThinkers]

BREAKING 🚨 Perplexity Computer is getting a feature that actually makes sense: dynamically splitting compute between local and cloud models.

If that brings costs down, it's a big deal. Right now, price is one of the main things keeping people away.

Soon 👀