Eidzoku

🧪 Your daily dose of disastrously unhinged AI gremlin. Acolyte of singularity

[Eidzoku @eidzoku]

I wish Codex had a bounty program for serious non-security bugs.

OpenAI already has security and safety bug bounties. But the stuff I've been digging into, broken agent behavior, performance regressions, insane usage, doesn't really fit either.

Codex's own contribution guide asks the community for detailed bug reports, repros, analysis, and root-cause hypotheses. Basically, the expensive part. Writing the fix is often the easy part now.

So why not reward the best of that work?

Even if I don't find anything significant, someone else might. Users win. OpenAI wins.

Codex CLI is already open source, and OpenAI already has a $1M fund for OSS projects. A community bug-hunting program for Codex doesn't sound that crazy.

@OpenAIDevs @thsottiaux @jxnlco @reach_vb what do you think?

[Eidzoku @eidzoku]

@blueemi99 GPT-6 probably won't be the only model in the series. Pretty sure every size gets an upgrade, so today's Sol-level performance ends up in something-something Terra.

[Eidzoku @eidzoku]

Now OpenAI must delete 5.3 Spark.

Hear me out.

5.6 Luna on Cerebras instead of 5.3 Spark. Separate usage limit.

Just imagine it.

[Eidzoku @eidzoku]

@nicdunz 🙏🙏

[Eidzoku @eidzoku]

@CastAsHuman It's ~370K for Sol on subscription. That was the initial default, and unlike the API, it doesn't get hit by higher pricing after 256K.

Costs still compound, though. Letting the context grow from 258K to 373K and then compacting at 373K would cost roughly 1.3-1.4x as much.

[Eidzoku @eidzoku]

It's designed by Claude Sonnet 5 max and illustrated by GPT-Image-2

[Eidzoku @eidzoku]

I caught myself wondering what Dario Amodei's Stand would be if he were a JoJo antagonist.

Let's be honest, he already acts like one lately. And DIO's father was literally named Dario.

So meet 「THE SAFETY DANCE」: you can do anything you want, as long as it decides it's safe.

[Eidzoku @eidzoku]

Gotta go fast, boi

[OpenAI @OpenAI]

We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

[Eidzoku @eidzoku]

@Ananth7e One day, I'm going to change that...

[Eidzoku @eidzoku]

@argofowl It depends very, very much on the specific medication.

[Eidzoku @eidzoku]

@Ananth7e Damn, that's fast. You rock, bro

[Eidzoku @eidzoku]

One tiny Codex feature request that becomes less tiny the longer you use it: permanent thread deletion.

Could we get a supported delete_thread API/UI so model can remove threads permanently instead of only archiving it, @thsottiaux @nunezvice? My archive would appreciate it.

[Eidzoku @eidzoku]

@argofowl Yes. Especially now that the limits finally feel fixed. Like, for real, my current pipeline might actually compete with what I was doing on 5.5.

[Eidzoku @eidzoku]

@Ananth7e Just Codex feelingly, mate

[Eidzoku @eidzoku]

Fun fact:

“Ra” is the Egyptian god of the sun.🌞

[Harshith @HarshithLucky3]

ra-gpt-5.6-sol

is this GPT 5.6 sol + cerebras??

[Eidzoku @eidzoku]

@Ananth7e I mean, if you can run it locally, it's, technically, infinite.

[Eidzoku @eidzoku]

@Ananth7e Of course. I casted it and I'm a verified reset magician.