Tiberium

I'm not sure why we are even arguing here, you can go into Codex or ChatGPT Work and start using 5.6. It'll start draining your weekly Codex limit. There's no unlimited 5.6 in Codex.

It's not Codex-specific, 5.6 is for chat. Look at another entry, for example:

"Expanded projects, tasks, and custom GPTs"

Do you have custom GPTs in Codex?

This is just the generic ChatGPT plan comparison, but under the Codex url.

I don't see that? It's unlimited in chat on the Pro plans, not in Codex.

> the forward, contract-driven line; master carries the pc64 x86-64/UEFI world

I have no idea what this means. Also, the spam of top-level AUDIT-*.md files doesn't inspire confidence :)

You're wrong for OpenAI, their 5x/20x always referred to the full weekly limit, and my (and others') multiple tests confirmed it empirically multiple times by calculating API-equivalent prices, especially when going from one subscription tier to another.

You're right for Anthropic though, their 5x/20x is mostly for 5h limit.

They're halving the $200 sub so it essentially becomes 10x from 20x

> 64 icons. Six rounds. One winner.

The winner is Claude

If anything, Spritefusion seems to be higher quality, especially when destroying text.

I don't really see evidence that it's an "OpenAI" model, the title is IMO very misleading.

Is this a whole blog post for a simple backend that just routes based on the User-Agent?

I honestly thought there would be some elaborate chain there, not "we forgot to use encryption"...

That's a great approach, I think byte matching is just popular because it's extremely easy to test in the end: are the bytes the same? While your approach requires putting far more trust into the tests.

The whole readme is extremely Claude-written, its dense Claudish style leaks from every sentence ;)

If anyone's curious - https://planetscale.com/docs/postgres/search/get-started#loc...:

They're not providing a local extension with the same performance at the time - it's only offered on their cloud services.

The local version https://github.com/planetscale/lead is mainly just for testing the syntax, it doesn't have the same perf characteristics.

I hope the future generations of models will stop with the fluff/textual clutter/overexplanations - I don't know the right term. They easily fix those once you tell them, but they still do them by default. What I mean on this page, for example:

> A WORLD, FOUR BILLION YEARS IN THE MAKING

> DEEP TIME · CONTINENTS IN MOTION

> Distance becomes an ocean.

> Continents drift apart. Shallow seas spread across land that will one day be dry.

> EXPLORE 4.54 BILLION YEARS OF EARTH HISTORY

> EARTH THROUGH TIME

> Scroll to travel through time Drag to orbit

> THE PAST IS ANOTHER WORLD. IT IS ALSO OURS.

> ILLUSTRATIVE TRANSITION

> Ga = billion years · Ma = million years / Event spacing is not linear

It gets to the point where if you ask a model to make a tiny toy "OS", it'll write stuff like "REAL MODE. REAL MULTITASKING. DRAG WINDOWS / TAB TO SWITCH" directly in the background of the desktop of said toy OS.

https://github.com/i2cjak/T3CAD seems like another promising project

Seems like pure SEO spam?

> The model has appeared on the ChatGPT API.

But it hasn't.

This is just C2PA for Claude-processed files, apparently. It's not related to their text watermarking.

C2PA is file metadata and can be trivially stripped away, unlike hidden watermarks, e.g. SynthID.

This isn't an open source project like Anubis and apparently you have to pay $50 for advanced features:

> POWBlock is shareware. Good shareware we hope, because we only gate the optional "enterprise-level" super-nerd stuff behind a paid license and give 100% of the main features for free. But for that reason the source code is closed and private. That being said, we're not scared of code reviews. If you're a bona fide security researcher willing to sign an NDA and publicly vouch for what you find, we can let you look at it.

> Reach out to us and/or drop some crypto. A license key is a flat 50 US dollars. If you buy as an individual operator, the key is valid for your lifetime for as many copies as you want, any where you want, just as long as you are the sole owner/renter of the physical servers you deploy on. Corporate/government entities need 1 license key for each US State and/or non-US country they plan to deploy it in, but can do unlimited deployments in any such place that they have a license.

I'm surprised that this distro's headline is "The Windows 7 experience on Linux" even though it's using a vague Windows-style XFCE theme.

If someone's interested in a real Windows 7 lookalike, take a look at https://gitgud.io/aeroshell/atp/aerothemeplasma, it's extremely close.

See for yourself:

- Kumander: https://www.kumander.org/img/images/kumander.png

- AeroShell: https://gitgud.io/aeroshell/atp/aerothemeplasma/-/raw/Plasma...

There is quite old research on this:

- https://arxiv.org/abs/2311.14455

- https://arxiv.org/abs/2401.05566

- https://arxiv.org/abs/2410.13722

I'm sorry, but it looks like someone's tiny game demo when they're learning HTML, with random mismatching assets pulled from some free packs (I guess in this case they're just generated). I fail to see any novelty or technical curiosity in this.

Unfortunately this is a very misleading post. I don't doubt the end result - it's useful that it works, but it's not "natively" and, unlike what some people assume, Claude didn't write a driver. It basically used HP's existing proprietary driver in a Linux VM on macOS, and just bridged that to macOS.

It also requires a root launcher that runs code from the user ~/.hp1008 dir, so security is weakened.

Unfortunately this is a very misleading article and headline. I don't doubt the end result - it's useful that it works, but it's not "natively" and, unlike what some people assume, Claude didn't write a driver. It basically used HP's existing proprietary driver in a Linux VM on macOS, and just bridged that to macOS.

It also requires a root launcher that runs code from the user ~/.hp1008 dir, so security is weakened.

So we can just make posts with any titles like that now? Look, I understand the optimism in open-weight models, but this headline is, well, far from reality. This repo has a lot of huge claims with a lot of amazing scores in the table, but zero proof.

As I understand, the code part is in https://github.com/Tiger3807861189/J-Space-Cognition-Suite-V... but there's nothing there that would make V4 Pro perform as well as Fable, it's mostly prompting - that's not even a proper implementation of "J-Space".

Isn't a lot of OpenAI infra also on Azure? They seem to be faring better than GitHub.

3.7 Flash gets 56 on AA up from 52 for 3.6 Flash. But it seems like this is at the cost of more output tokens per task: 3.6 Flash is 26k, 3.7 Flash is 37k. Due to 3.7 Flash's 2x slashed pricing it's still cheaper per task.

Unfortunately in this case that's not true at all, there was no provider with genuinely close or the same effective prices (mostly based on cache hit cost) to old v4 flash or v4 pro. People have this misconception that other providers must be much cheaper than the official one in case of open weight models.

If you check on OpenRouter, some other providers serve V4 Flash at seemingly cheaper normal input/output tokens rates, but with a huge caveat: they have at least a 5x increase of the cache hit cost of the official API, some have a 10x+. No provider comes close to Deepseek's old low cache prices, and cache is 90%+ of what matters in agentic sessions.

Closest comparison:

- Deepseek: $0.14/$0.28 with $0.0028 cache hit cost for official API

- DeepInfra: $0.08/$0.18 (cheaper base rates!) with $0.016 cache hit (almost 6x!! Deepseek's current cache cost)

Another great example is Kimi K3, official API is $3/$15 and the cheapest provider on OpenRouter is $2.8/$14, only a tiny difference.

Funny result from the web demo. I'm well aware that it's an extremely small and, well, stupid, model, but even so:

Query: HN

Result:

{ "function_calls": [ { "name": "lock_door", "arguments": { "door": "front door" } } ], "reasoning": "User wants to lock the door. No specific door mentioned, so use 'front door' as default.", "confidence": 0 }

I'd expect it to at least ignore (call no tools) for the queries that it doesn't understand. And it seems like it does do that, just not consistently.