Eidzoku

🧪 Your daily dose of disastrously unhinged AI gremlin.

Room-temperature takes, irony, sudden puns & deep thinking.

FOLLOW ME! Embrace the chaos! 📈

[Eidzoku @evi77ain]

You. Yeah, the one who sees this post.

Are we friends yet? No?

Why friend-shaped then?

[Eidzoku @evi77ain]

@LexnLin Can't fly without a reset...

[Eidzoku @evi77ain]

@Ananthr27104587 I bet the thing that made them float is the speed of their limits draining.

[Eidzoku @evi77ain]

@DeryaTR_ We're about to witness such a great time.

[Eidzoku @evi77ain]

@derrickcchoi Fly me to the Moon🎶

[Eidzoku @evi77ain]

@rezoundous It's not like it's drastically worse imo

Anyway, 5.5 is underperforming lately.

[Eidzoku @evi77ain]

@Kappaemme1926 You wanted to say 'to use codex from a few weeks ago' cuz 5.5 is a bit silly goose atm!

[Eidzoku @evi77ain]

@birdabo I'm pretty sure they're too short on compute.

I mean, what's the point of shutting Sora down just to release another video model a few months later, lmao.

I bet they haven't given up on this idea, but it's not going to happen anytime soon.

[Eidzoku @evi77ain]

@juleslogs Hmm. I think it's having a real discussion about something.

Not like people never argue, obviously.

I mean the real one.

The kind that lets either you or the other person learn something new.

A constructive kind.

[Eidzoku @evi77ain]

@argofowl The only thing flying rn is my quota consumption tbh...

[Eidzoku @evi77ain]

Universal level of vagueposting

[OpenAI @OpenAI]

It's time to fly.

[Eidzoku @evi77ain]

@OpenAI I mean, they were floating, not flying...

[Eidzoku @evi77ain]

@brockpierson I'm Eiji.
Mastered the art of having fun wherever I am at a given moment.

Including, but not limited to: silly puns, shitposts, AI autism, and thinking about something deep and beautiful.

Also, I spend a hell of a lot of my time building things for, like, three people. Join me!

[Eidzoku @evi77ain]

@synthwavedd GLM

[Eidzoku @evi77ain]

@brockpierson But where's the line between small and big account?

[Eidzoku @evi77ain]

The question is still hanging around.

I know it's most likely under NDA for now, but maybe our prayers will be heard and we'll receive an answer, @thsottiaux

Pretty please 🥺

[Eidzoku @evi77ain]

Real concern: are limits of ChatGPT and Codex gonna be merged as well?

[Eidzoku @evi77ain]

@shikhr_ I bet they were optimizing for tokenmaxxing before, and people used their budget to generate random bullshit just to look productive.

[Eidzoku @evi77ain]

@mark_k @ajambrosino PowerShell by itself is goofy as hell.

Also, even if it's supported, I bet most LLM training data is still Linux-native or so.

That means we inevitably lose quality and speed.

[Eidzoku @evi77ain]

Microsoft be like:

[Mustafa Suleyman @mustafasuleyman]

Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier.
First is our text foundation model, MAI-Thinking-1, exceptionally strong on reasoning and SWE tasks.
- It’s a 35B active parameter MoE with a 256K context window. Independent human raters on Surge prefer it for overall quality in blind side-by-sides versus Sonnet 4.6, and it’s achieved 97% on AIME 2025, the key measure of its general-purpose reasoning abilities.
- It's at 53% on SWE Bench Pro, placing it right alongside Opus 4.6 on one of the toughest coding benchmarks.
- And since we co-designed our models with our own silicon, MAI-Thinking-1 is optimized on our MAIA 200 chip. Benchmarking head-to-head against the GB200, we see 30% better performance per dollar as well as a 1.4x performance-per-watt gain when running our MAI models on the MAIA 200 end-to-end.

Next is MAI-Image-2.5 and its Flash variant. Two super strong models now at #2 on the leaderboards, surpassing the score of Nano Banana 2 on image editing.

Last for now is MAI-Code-1-Flash, our new inference efficient coding model, especially tuned for VS Code and GitHub Copilot CLI.
- Code-1-Flash achieves 51% on SWE Bench Pro, despite having just 5B parameters, putting it closer to Haiku in size but cheaper in cost.

All of this is the foundation for Microsoft Frontier Tuning. It lets you customize our models to create custom, company-specific agents that only you control. You can make our model, your model. Your data. Your agents. Your moat.

Early adopters are already seeing a difference. When we tuned our models for McKinsey’s tasks, MAI delivered the highest win rate, outperforming GPT-5.5 on quality, while being 10x lower on cost.

Also really excited to be collaborating with the amazing team at Mayo Clinic to jointly train a new frontier AI model for healthcare.

Our announcements today mark another milestone on the road to humanist superintelligence. You can learn more and about our other new models in our latest blog: https://t.co/v65eop5Ixq