lucamark

Exactly! And they should also release the new inference engine in this month. Anyway, I am curious to try Opus 5, considering that previous versions (e.g., 4.8) were disappointing

But why GPT 5.6 Sol is so behind on the benchmarks? In real-world projects, it is the best frontier model to me in terms of accuracy, speed and consistency. It can just be compared to Fable 5, but I prefer GPT 5.6 Sol because of inference speed.

I've never trusted on model cards though. I'm sorry.

A month ago I posted Modeloop here, a visual environment for designing control systems (https://news.ycombinator.com/item?id=48543968).

In the comments,boscillator wrote:

"I'm always saying Simulink needs a text-based interface. Same signal flow programming model that supports scopes and continues time integrators, just with text instead of drag-and-drop." (https://news.ycombinator.com/item?id=48598291)

This has been on my own list for a long time too, so I built it. It's called MDL (Modeloop Description Language).

Every diagram now has an equivalent textual form, and the two stay synchronized in both directions. This allows users to decide if to create and modify diagrams as blocks on the canvas or as MDL text.

Curious what HN thinks — and boscillator, if you're around: this one's on you.

For firmware RE this would be most useful if every guess is tied to evidence

Using power as feedback is neat because it works when you can’t instrument the target but I wonder how they separate new behavior from boring analog noise, e.g. the clocks

I’d like to see which parts of the assurance case are language/toolchain dependent versus plain old process, tests and traceability.

Remember that such models are available to us thanks to NVIDIA GPUs

Never had rejections in the short time Fable was available

Agree. It is a mediocre model, expensive while not being a frontier

Wrong! Look at it better. It shows that Opus has superior performance but at higher cost.

You're referring to the Agentic search, but if you look at the Agentic computer use the cost is basically halved.

However, I am also confused about market positioning. Too expensive to perform daily tasks - open souce models are much cheaper - and not frontier model to address complex real world problems.

Rarely used Sonnet btw.

I checked before posting but I didn't find it

OpenRouter?

> In other words, the Trump administration appears to be pressuring OpenAI to do what Anthropic is already voluntarily doing: keeping its most powerful AI models under wraps.

while I generally agree that models with this level of cyber capability should not necessarily be released to everyone on day one, if a company or developer can meet compliance requirements, they should be eligible - and eligibility should not be based on nationality.

Your point is interesting! And with a Mythos-class model already on the market (e.g. DeepSeek or GLM), will OpenAI and Anthropic be forced to apply the same policy?

Uh great! I haven't seen these posts, thanks for the links :)

Expections occur

In many other european countries, the story doesn't change much - with some exceptions. The irony is that the EU is now discussing EU Inc. a law to let startups incorporate online within 48 hours, for at most €100, under a single optional corporate framework across Europe.

That is exactly the right direction. But for now it is just a proposal

Do you think that stuffs are different in Italy, Spain or France? Just to cite some others

Europe will suffocate under the weight of its own bureaucracy. The sad thing is that it is not new to me, I've heard so many stories like this one. This is the kind of friction that makes founders incorporate elsewhere.

Also, a founder spending months coordinating lawyers banks and tax advisors is not talking to customers or building the product. The opportunity cost here is huge.

Anyway, you are pretty close. One more push, don’t give up. :)

That's bad for ETF owners too. Funds will be legally forced to buy a tightly restricted supply that will pump the price even more

Let's grab some popcorn.

- Price/Sales about 128x (NVIDIA had a peak of a 40x at its peak)

- Bought Twitter per 44 billions. Inflated its valuation to $250 billion just by integrating it into X.AI

- EnterpriseValue/EBITDA about 219x (30x for scaleup business) and negative Price-to-Earnings

- Low free-float trick (minimal public shares available)

Even the market efficiency hypothesis struggles to justify it

I personally found this NVIDIA move very interesting. Automakers generally do not want to become frontier AI infrastructure companies and they love technology standardization.

The real technical challenge is rappresented by edge cases: a software that is excellent 99.9% of the time can still be unacceptable if the remaining 0.1% contains rare but catastrophic scenarios. And that's why we still don't see many self-driving vehicles on the roads today.

However, NVIDIA has a credible shot because it controls much of the loop - hardware, training infrastructure and simulation environment. If it works they will impose a huge vendor lock-in, difficult to replicate for other competitors.

For sure! Thanks again for noticing it and good luck with your projects as well :)

Good point! Today Modeloop mostly treats signals as numeric values with optional metadata.

I probably wouldn’t want Modeloop to automatically convert between SI and Imperial units inside the model, because that can make the behavior less explicit.

What I do think is useful is documenting that a certain signal represents speed in mph, temperature in degC, pressure in psi, etc. So the value stays exactly what the model defines, but the user and generated artifacts have clearer semantics about what each signal represents.

First of all thanks for trying the product and sharing your feedback! Really glad you liked the onboarding :)

I think the closest term is probably contextual onboarding or guided onboarding. This onboarding system is created around concrete real-world examples so users don't feel disoriented. The user starts from a working model and then the UI highlights the next meaningful step.

Technically it is a pretty minimal implementation without using external libraries - I generally avoid the use of libraries when possible. Given an example model, define a sequence of expected actions, store the current onboarding step in app state, and render a hint/highlight near the next control until the user completes that action. Then just optimize the transitions timing and appearance

Makes sense. By the way, the Modeloop UI is clean and standard

Totally agree with you! Just to be transparent—I like to be—this project was launched less than a month ago. It hasn't had an audience yet, which is why I haven't spent much time on the landing page.

For example, also the bouncing ball does not catch the real essense of Modeloop. Indeed, I will change it soon with a more control-oriented example :)

Edit: Changed the example on the landing page with a DC Motor controller

To be legally binding and enforceable, they must be explicitly and individually approved by the user. Even when signing in as a guest/without an account, using the app creates a license agreement, so the clickwrap flow must legally capture this specific assent to protect the underlying engine's IP.

Regarding the needs a text-based interface, I totally agree with you! Most of the time you can create models directly within the terminal or with a simple script. That's why we actually provide a CLI that allows you to build and run models. You can initialize models, add blocks with parameters, connect ports, validate the graph, and trigger code generation all from the terminal. Nevertheless, we think the GUI is useful to review the model and inspect it graphically

Uuh great! I will be honored to hear your feedback and that of your team. Just to clarify, the main users we see today are robotics, embedded, and control systems engineers. That said, I can also see it being useful for solution engineers working closely with customer-specific systems. I’d really value your feedback after you’ve tried it.

Please use the HN_PRO_2026 promo code to unlock more features :)