Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT-5.6 Sol Ultra will be in Codex

twitter.com

Post reply on HN
Print view View original

311–320 of 433 posts

Re: GPT-5.6 Sol Ultra will be in Codex

#311

Post by kzrdude »

alightsoul wrote: post #294
schnitzelstoat wrote: post #202

Earlier quoted context omitted.

It gives them a massive advantage because they can cut the cost per token by a lot and eat Anthropic's market share. In what universe is any company going to give that advantage away? In any case if they take away a lot of market share it's basically the same in the end - most people will be using these optimisations.

Deepseek gave it away with R1. They did it again with V4 and DSpark. You don't use it because it's Chinese

It's available from opencode (non-chinese hosting), additionally deepseek v4-flash is available with rate limits of course, for free, there.

Re: GPT-5.6 Sol Ultra will be in Codex

#313

Post by kzrdude »

timcobb wrote: post #31
OutOfHere wrote: post #21

Earlier quoted context omitted.

[flagged]

are you, by any chance, a bot?

Unfortunately, Firefox is often put in the bot bucket nowadays. Not sure what happened here. It might be because of aggressive privacy settings that are flagged as unusual or suspicious.

Re: GPT-5.6 Sol Ultra will be in Codex

#314

Post by adrian_b »

reinitctxoffset wrote: post #172
satvikpendem wrote: post #159

Earlier quoted context omitted.

The US can just ban Chinese weights being used in US companies.

If the Trump administration decides to annoint Altman and Amodei in defiance of market forces it will rapidly discover that it no longer has the sovereign bond auction pricing power to prop them up. This isn't 1998: the Treasury has taken five major body blows in the last 25 years, the world's energy markets, maritime insurance regimes, electronic payments rails, and moral authority in places like the UN Security Cou…

NVIDIA appears to play on several parallel paths that may compete with each other.

On one hand it appears to cooperate with OpenAI and Anthropic, as big customers.

On the other hand NVIDIA cooperates with Palantir, providing the HW for its "Sovereign AI OS" (a turnkey system including HW and SW for local inference and post-training/fine tuning) which uses the slogan "The future of AI is on-prem" (i.e. not as a customer of OpenAI or Anthropic, but using an open-weights LLM, e.g. a fine-tuned NVIDIA Nemotron or a Chinese LLM).

Presumably with the goal of promoting their competing solution, Alex Karp (Palantir CEO) has delivered a few weeks ago a very harsh criticism of Anthropic and OpenAI (who allegedly inflate the token consumption and they might also steal the data of their customers, which must be sent to them).

So NVIDIA both cooperates and competes with OpenAI and Anthropic.

Re: GPT-5.6 Sol Ultra will be in Codex

#315

Post by cactusplant7374 »

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…

That's strange. One can easily steer their session to use agents proactively.

Re: GPT-5.6 Sol Ultra will be in Codex

#316

Post by JeremyNT »

throw394042 wrote: post #30

I'm working in large US corporation. And I see that I already have access to 5.6-Sol Ultra on my corporate account. I haven't really used it yet. 2 months ago management was showing us scoreboards, praising leaders who used most tokens. Last few weeks, we're getting weekly emails, telling us that whenever we can - we should use cheaper models, and that we should watch the page which shows our tokens usage.

> 2 months ago management was showing us scoreboards, praising leaders who used most tokens. Last few weeks, we're getting weekly emails, telling us that whenever we can - we should use cheaper models, and that we should watch the page which shows our tokens usage.

GPT 5.5's double token cost was the threshold for me. These things are getting expensive quickly - the subsidized pricing can't go on forever.

Re: GPT-5.6 Sol Ultra will be in Codex

#317

Post by catigula »

karahime wrote: post #263
rowanG077 wrote: post #241

Earlier quoted context omitted.

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…

I don't know of any modern workflows that rely on "we'll tell the person not to do it again", though. There's a reason that companies have adopted blameless postmortems, because if your response to the DB going down is "It's fine, Kevin learns and next time he won't misuse the prod credentials", you are guaranteeing prod will go down again in the same way at some point.

Every modern workflow implicitly relies on that. No infrastructure is fully robust. There's a senior DB person who has learned many things many times over who could bring down most of the US power grid.

Re: GPT-5.6 Sol Ultra will be in Codex

#318

Post by rowanG077 »

karahime wrote: post #263
rowanG077 wrote: post #241

Earlier quoted context omitted.

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…

I don't know of any modern workflows that rely on "we'll tell the person not to do it again", though. There's a reason that companies have adopted blameless postmortems, because if your response to the DB going down is "It's fine, Kevin learns and next time he won't misuse the prod credentials", you are guaranteeing prod will go down again in the same way at some point.

Literally the entirety of the worlds infrastructure relies on that. In the past we had (literally) had nuclear war hinging on a single person just deciding that some data point is an artifact.

Re: GPT-5.6 Sol Ultra will be in Codex

#319

Post by rowanG077 »

benjiro29 wrote: post #260
rowanG077 wrote: post #241

Earlier quoted context omitted.

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…

> 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 entire problem that you don't even have access to models if you use a provider like anthropic/openai

Re: GPT-5.6 Sol Ultra will be in Codex

#320

Post by cactusplant7374 »

Szpadel wrote: post #201
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…

sources for easy confirmation: https://github.com/openai/codex/blob/98d28aab54ed86714901b66... https://github.com/openai/codex/blob/98d28aab54ed86714901b66... https://github.com/openai/codex/blob/98d28aab54ed86714901b66...

Very strange because in the TerminalBench benchmark Ultra does better than Sol. They didn’t add the reasoning level to the chart.
Post reply on HN