I don't think so? I think OpenAI had planned to half the limits but decided not to go ahead with it, but this analysis was done at the limits we have now. I could be wrong about this though?

> Opus 5.5 costs ~6x of Sol 6.1.

My point in part is that you can't compare like this.

Opus 5.5 is a more capable coding model than Sol 6.1. AA benchmarks show Opus 5.5 max effort scoring 62 vs Sol 6.1 max effort on 54: https://artificialanalysis.ai/agents/coding-agents?agents=de...

Unfortunately AA doesn't report non-Max efforts for their coding benchmarks.

But no one runs them on max. For Opus the advice is to usually run it on medium effort, and I've found that advice to be good.

For me to get equivalent results on Sol I need to run it on high effort, and I need to use Astra as an orchestrator (whereas Opus medium does great as an orchestrator too).

That combination pushes the blended token pricing up a lot.

I can't give you hard numbers on this, but at the moment my Codex credits get used up a lot quicker than my Claude credits on similar tasks in the same codebase.

I'm glad we have two great models competing, but at the moment I'd say Opus has the edge.

There was some chatter on Twitter, Reddit and here when an earlier version of this came out where people normalized it by completed AA or Epoch task and Claude models still came out well ahead.

In practice OpenAI subscriptions are more competitive than they appear because OpenAI does a lot more subscription resets. But that's hard to account for!

Personally, for coding tasks I've found Sol 6.1 High is about as good as Opus 5.5 Medium for well defined tasks, but Opus is much better at orchestrating.

I have both subscriptions, and when using Claude I use Opus Medium to orchestrate other Opus Medium subagents, whereas on Codex I use Astra Medium to orchestrate Sol High subagents.

They absolutely have said that.

> That's why the prices of frontier models have been skyrocketing,

They haven't.

"The cost of achieving a given level of AI performance has fallen about 47% per quarter since 2023, or 13× per year."

"The cost of a given level of performance often falls fastest right after that level is first achieved, that is, when it is state of the art (SOTA). We see this pattern on three of our five main benchmarks of AI capability. Averaging across all five, cost falls 66% per quarter (75× per year) for performance that has just debuted as SOTA."

https://epoch.ai/publications/the-plunging-price-of-thought#...

On a per-token basis Fable 5.2 is cheaper than 5.1:

5.2: $8 per million input tokens and $40 per million output tokens.

5.1: $10 per million input tokens and $50 per million output tokens.

Compared to older (and much less capable) frontier models it is also cheaper. For example GPT-4 was $30 per million input tokens and $60 per million output tokens

https://developers.openai.com/api/docs/models/gpt-4

Be very careful comparing the Artificial Analysis "cost per task" scores. Fable 5.2 is expensive there because it is doing tasks the other models fail at.

And for most people here who care about coding performance Anthropic has a substantial lead[1] - although sadly they don't have coding benchmarks separately for Opus 5.5 Medium which I think it the sweet spot.

The cost tab really should have something like "how many tasks can you do in a particular subscription tier" for many people. I realize this changes often though.

[1] https://artificialanalysis.ai/?models=gpt-6-astra-high%2Cgpt...

> They essentially failed at that, assuming they kept their training costs for the model step at least constant.

What? No!

This is completely wrong!

This proves that inference scaling continues to work beyond the previous frontier.

Claiming that higher inference costs shows failure is an odd thing to think. We've seen this optimization cycle over and over, most recently with Opus 5 -> Opus 5.5

Opus 5 was slow and expensive. Opus 5.5 came out and was better, fast and much cheaper for them to run.

There is nothing to indicate the same can't happen here.

Anthropic probably has multiple options open to them with this model, and they'll probably explore all of them:

- Use collected Fable 5.2 traces to improve reasoning economy for Fable 5.3

- Use Fable 5.2 as a teacher model for Opus 5.6

- Add additional fields RL data to expand Fable capabilities for Fable 5.4

It's not exactly confidential that NVidia wants to sell more chips in China. They have publicly lobbied for it, and personally I think the evidence is that their arguments make a lot of sense.

The affair was dumb etc but from a business point of view it seems harmless.

From a geopolitical point of view one could me mischievous and argue that Jensen seduced a Chinese-linked tech executive and persuaded her to buy American tech instead of supporting domestic Chinese options. Good work Jensen!

It's like saying "developers should be responsible for their own security so we will make them liable if a random node dependency in the application they are building gets pwned"

Firstly, liability and security are different things.

Employees should never be liable for something that happens if they are using a work-supplied tool at work in the manner in which it is designed to be used.

Secondly, it's just lazy thinking to say "oh it's hard therefore we'll make the human sorely responsible".

There are many things that can help - sandboxing, adversarial LLM inspection of prompts, separation of security contexts etc. Yes they are hard and might need new primitives.

For the second: No, not fine tuning at all.

For one thing in-context learning is enough for learning novel tasks[1].

Secondly, it can do new things outside that data. LLMs build models of the world[2]. That means it can model novel behavior based on that internal representation.

[1] https://arxiv.org/abs/2005.14165

[2] https://arxiv.org/abs/2305.11169

I can't find it now, but somewhere off https://shorepine.github.io/amy/piano.html it talked about the Android work you did.

Edit: There is https://github.com/shorepine/alles#thank-you-to but I remember a more detailed thing too.

I always notice your name because I enjoyed Advogato back in the day and keep using bits of your work elsewhere.

A neural network can do new things outside the training data. Random initialization sees to that.

Whether those things are useful is a different question, but it turns out that yes, some of those things are useful. It also turns out that in a large-enough LLM if you give it some kind of idea of what you see as useful then yes, it can pick out the newly useful things.

This isn't true.

While the "copyrightability" report[1] mostly deals with images (it was written before coding agents) it's quite clear how code generated by an AI driven by humans can be copyrightable:

> The Office finds that the compilation of these images and text throughout the Work contains sufficient creativity under Feist to be protected by copyright. Specifically, the Office finds the Work is the product of creative choices with respect to the selection of the images that make up the Work and the placement and arrangement of the images and text on each of the Work’s pages.

and

> Unlike prompts alone, these tools can enable the user to control the selection and placement of individual creative elements. Whether such modifications rise to the minimum standard of originality required under Feist will depend on a case-by-case determination. In those cases where they do, the output should be copyrightable

[1] https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...

It's more interesting on an open road but horrible to drive in traffic. Overall both of us hate the experience.

I get my new electric car (a Geely) on Wednesday.

Both things can be true.

I'm old enough to have lived without the internet. I'll take this over that anytime.

> Facebook has shown it cares more for money than being a steward. Don't see how you can trust them.

It's precisely this love of money that make me trust them. They value what they get from my data, so they do their best to keep it to themselves.

That's much better than ISP (for example) that sell raw traffic data.

Does this mean you recommend not using one? Because that goes against all the common best practice recommendations from when I did a couple of years working in Web3 stuff a few years back.

> I think it's naive to suggest that self-custody is just a novelty with no value.

That's not what I suggested.

I suggested that it is less secure over time than centralized exchanges, even with the things you listed.

Convenient date to pick there.

> which of those assets would you have preferred to own and why?

NVDA.

It's a productive use of resources, not a last-fool-standing pyramid scheme.

Given things like the ledger hardware attack[1] I think the time indexed risk of using a centralized exchange is probably lower than doing self-custody.

I understand that many people do self-custody for philosophical reasons, but I think trying to disguise this as a security decision ("If your crypto was stolen in this, it wasn't your crypto.") is arguable at best.

[1] https://www.ft.com/content/70f69253-6017-4259-95c9-48bf0e088...