GPT-5.6 Sol Ultra will be in Codex • Hacker News Board

Re: GPT-5.6 Sol Ultra will be in Codex

#401

Post by djsjajah » Tue, Jul 07, 2026, 1:01 AM UTC

whstl wrote: post #394

RazorBucksICO wrote: post #380

Earlier quoted context omitted.

“Asshole non-technical people” You mean the people creating actual business value in every industry where software isn’t the product? Speak not unless spoken to, Mr. cost-center!

Without the "some" qualifier my quote totally loses meaning. In the same message you managed to twist my words, lack reading comprehension, jump to conclusions and do a personal attack. Since you're new here, I recommend reading this: https://news.ycombinator.com/newsguidelines.html , especially this part: "Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier t…

I think they were making a joke. In the future, you might consider the advice you are giving as well as giving it.

Re: GPT-5.6 Sol Ultra will be in Codex

#402

Post by satvikpendem » Tue, Jul 07, 2026, 1:28 AM UTC

gorszon wrote: post #226

satvikpendem wrote: post #160

Earlier quoted context omitted.

It is not because they want to but because they literally don't have the capacity to.

What a suprise, to make better models, they just making them larger and larger, then they can barely run it, after a round of LARP-ing that they invented some dangerous LLM?

Yes, that's how you get to the quality of Fable versus worse models these days.

Re: GPT-5.6 Sol Ultra will be in Codex

#403

Post by satvikpendem » Tue, Jul 07, 2026, 1:28 AM UTC

zaptrem wrote: post #205

satvikpendem wrote: post #160

Earlier quoted context omitted.

It is not because they want to but because they literally don't have the capacity to.

Why haven’t we seen any queues or the like over the past week then? If it’s truly a capacity limitation why not just boot subscription users to a lower priority queue or limit usage to outside peak hours?

They are, that's why Fable is going away tomorrow for example.

Re: GPT-5.6 Sol Ultra will be in Codex

#404

Post by satvikpendem » Tue, Jul 07, 2026, 1:29 AM UTC

alightsoul wrote: post #390

satvikpendem wrote: post #378

Earlier quoted context omitted.

Misleading first sentence, as most use OpenAI and Anthropic directly via their own LLM not through OpenRouter. But you're right that lots of companies use open weight and Chinese models, not sure how the parent got to their conclusion.

because in the us you get plenty of people who refuse to buy non western products solely because they are not western. like chinese cars which were banned.

A small subset of the population not buying non Western is not indicative of the larger trend.

Re: GPT-5.6 Sol Ultra will be in Codex

#405

Post by gck1 » Tue, Jul 07, 2026, 3:15 AM UTC

behnamoh wrote: post #14

I still don't know why OpenAI doesn't put gpt-5.5-pro in Codex. It's one hell of a model and easily parallels Fable/Mythos. Sure, it'll use up your quota much faster but that's the price some users are willing to pay for absolutely high quality responses. I think gpt-5.5-pro runs 12x parallel gpt-5.5 agents behind the scene and uses OpenAI's secret sauce to synthesize their answers into one insanely good response.

Forget 5.5-pro. Why isn't everyone talking about the fact that there's no 1M context window model in codex?

Yes, I know attention degrades above ~200k, but it's still useful in many applications.

Re: GPT-5.6 Sol Ultra will be in Codex

#406

Post by killix » Tue, Jul 07, 2026, 4:21 AM UTC

d4rkp4ttern wrote: post #259

Szpadel wrote: post #199

there seems to be very big misunderstanding about what the "ultra" is, so let me explain it basing on the codex source code: it's similar to Claude code ultracode. there is no ultra effort level implemented on the backend. it's just alias in the codex to max effort setting and single line addition to prompt to use subagents proactively. that's all as far as we know pro models work differently. for once those are back…

> single line addition to prompt to use subagents proactively. This misses an important detail. In Claude Code [1], ultracode suggests the agent create a JavaScript code to deterministically orchestrate sub agents. This is different from just having the main agent launch sub agents and (non-deterministically) manage them. The resulting workflow is called “dynamic” because CC creates this orchestration script dynamica…

[flagged]

Re: GPT-5.6 Sol Ultra will be in Codex

#407

Post by solenoid0937 » Tue, Jul 07, 2026, 4:53 AM UTC

alightsoul wrote: post #391

solenoid0937 wrote: post #333

Earlier quoted context omitted.

I run local models every day. We are in a race to superintelligence. The first country to AGI will be the first to superintelligence, and the first to superintelligence will have de facto control over the world and the future of humanity. They will also be able to prevent others from reaching superintelligence. Of course it's in the US's best interest to slow down China. You aren't zooming out and looking at the big…

while china goes country to country organizing fairs through their embassies to get new business partners, the us sees the world as a zero sum game.

Global power does not run on kumbaya vibes. Superintelligence is a zero sum game whether you understand why it is or not.

I'm glad you trust a country that literally imprisons, rapes, and murders their human rights lawyers though. And one in which you cannot even criticize the government without being disappeared. Surely such a country is best positioned to have total control over all humanity, forever.

Re: GPT-5.6 Sol Ultra will be in Codex

#408

Post by throw1234567891 » Tue, Jul 07, 2026, 5:28 AM UTC

rowanG077 wrote: post #241

barrkel wrote: post #186

Earlier quoted context omitted.

People are stochastic. You build reliable processes out of unreliable parts with feedback and self-correcting mechanisms. AI is not actually magically special in this regard. It has higher variance and we're still figuring out how to get all the tradeoffs right.

The big problem is that a person making a mistake can be taught to not make that mistake again. That's also not foolproof but at least it works a lot of the times. AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck. That is to say I do agree that building reliable processes out of unreliable parts with feedback is the modus operandi. However A…

> The big problem is that a person making a mistake can be taught to not make that mistake again. That's also not foolproof but at least it works a lot of the times.

You can do the same to an LLM with the same outcome.

Re: GPT-5.6 Sol Ultra will be in Codex

#409

Post by throw1234567891 » Tue, Jul 07, 2026, 5:29 AM UTC

rowanG077 wrote: post #319

benjiro29 wrote: post #260

Earlier quoted context omitted.

> AI are unteachable, if you have given them a good prompt and they do something wrong 90% of the time you are shit out of luck. If the Model makes repeated mistakes on the same subject matter, you can update your agent.md file, or you can add skills to deal with specific prompts, or you provide a better default harness. The whole idea of coding agents is their harness makes a big difference vs a pure raw model. > Ho…

> If the Model makes repeated mistakes on the same subject matter, you can update your agent.md file ... That's all just prompting. > How do you think models are created? They are trained on feedback and learn. No one is post training models on a single mistake. At least I have not seen it. I also doubt it is effective. Post-training on a single failure will not meaningfully change the model. That even sidesteps the…

> That's all just prompting.

>> Joe, how many times have I asked you to pay attention to commas. Come on, "let's eat grandma".

Re: GPT-5.6 Sol Ultra will be in Codex

#410

Post by rowanG077 » Tue, Jul 07, 2026, 6:59 AM UTC

SR2Z wrote: post #392

rowanG077 wrote: post #386

Earlier quoted context omitted.

They're different mechanisms. Teaching modifies the learner. Prompting doesn't modify the model. It provides additional context that influences a single inference. A person who has learned something can apply it years later without being reminded. An LLM generally cannot unless the knowledge is incorporated into the model itself or provided again.

Right, but "provided again" is what SKILL.md or whatever else are for. The value of LLMs is that they're stateless. With sufficiently detailed documentation and a well-bounded task, they are quite useful.

I don't think it's generally thought of an advantage that LLMs are stateless. In deployment it's nice, it would be much better if the model could continually learn.

I'm not claiming LLMs are not useful, they most certainly are.