Reubend

> rendering DirectX 8 → DXVK → Vulkan → MoltenVK → Metal

Am I reading that right? It makes API calls that go through 5 different layers before actually getting rendered? That's kind of crazy. I'm surprised it works, although I guess the underlying libraries are solid enough that it shouldn't be unexpected.

I'm sorry, but I agree with the wiki editors in this case. Odin is obscure. The author of this blog post seems to think it's well known, but I don't think that's substantiated.

That understates how difficult it is to get to the level of performance they attained. The fact that it surpasses DeepSeek v4 in most ways shows that they accomplished some great work in this space.

The example images look great. Always good to have more diverse OSS models.

I think you're missing the point. Everything you said is theoretically correct, but the parent comment was talking about the concrete circumstance of pentesting with the top models today.

Let's just take GPT 5.5 and Opus 4.8 as an example. Both are worse than Mythos 5, but they're capable of quite a bit when the guardrails are lifted and they're paired with a skilled human operator. They more than "good enough" to reach the same result with the addition of some human effort.

It is strange, huh? But the hype cycles around these models often ignore good contenders. Xiaomi's MiMo-V2.5 Pro was doing really well and didn't get much hype either.

> I opened 120+ issues in Dgraph, Typedb and surrealdb looking for the perfect graphDB

Can you share some examples? What was wrong with those?

I don't really understand the connection; they went from image generation to medical scanning?

the WAV file decoding perf improvement is also very welcome!

> It’s literally impossible to catch Pac-Man?

I caught him! So it's definitely possible. You move a little bit faster than he does so you need to anticipate where he's going.

Seems like there's no official blog post with benchmark results yet. But I'm once again thankful for the Chinese AI labs for being open with their work and contributing it to the world under permissive licenses like this. The Fable 5 fiasco is just another reminder of how valuable these things are to have.

I've never understood the appeal of minimal writing apps. They've got beautiful UIs, but I could theoretically do everything here on a normal writing app, couldn't I?

Thanks so much for explaining this. I was just about to comment "Isn't Vulkan the main OpenCL and CUDA alternative these days?" could you share a bit more on what Vulkan is missing relative to the other two?

Yeah I was wondering about this too. It seems like way too much per GPU considering the purchase price of a B200 was around $40k last time I checked. So if we naively ignore the price of electricity and maintanence, it would only take 5 months before renting is a worse deal than buying outright.

> A University System Went All In on A.I. Now It’s Tearing Itself Apart.

The NYT often has a bias against AI, but the article's contents are actually a pretty reasonable summary of the different attitudes towards AI in academia. Then they went and slapped a terribly sensationalist headline on it, which doesn't seem well supported by the actual article.

Super cool side channel attack. I tend to agree that it's pretty impractical, but it's such a fun discovery!

> 0 ms Startup time

Is that true? It just goes right into the code with no initialization of any other libraries needed?

Hey there, I'm going to check out your project because the comments here have me a little worried that OP's project might have some quality issues.

Two things I found a little confusing from the docs though:

I couldn't easily find a page describing what it can't do yet. I saw that it only works with a "strict, deterministic subset of TypeScript", but is there a page showing what's included and not included in that subset?

Also, what's an "ambient surface" in this context? Is that a compiler term I'm just not familiar with?

Yeah that seems off to me too. But I guess they meant that since CockroachDB is compatible with Pg, it would also serve the same prupose?

This is such a refreshing policy. AI code is welcome as long as it's good, but comments have to be human.

If someone can't take the time to write their own replies (in their own words), then it feels fair to assume that they didn't take the time to test, review, and clean whatever code they submitted.

Yeah I really agree. There's no overage pricing on their website's main pricing page, which makes me think the jump from "Team" to "Pro" to negotiating an enterprise contract will really hurt. Going from $39/month to $199/moth because you needed slightly more is a really big jump in pricing. It's pretty much the opposite of what I would expect from a service that lets you scale to 0.

Props to them. That makes DeepSeek v4 Pro extremely cheap compared to others, even in the same category. Look at these prices per million outputs tokens:

DeepSeek V4 Pro: $0.87

Qwen 3.7 Max: $7.50

Grok 4.3: $2.50

GLM 1.5: $3.08

Opus 4.7: $25.00

GPT-5.5: $30.00

Seems like a branding mistake to me. This new device doesn't target the same group of customers that the Flipper Zero did, so it will be much harder for them to market it effectively, since the first device is already famous enough that people know them for that market.

What's the best remaining alternative?

Stupid question: I was under the impression that object storage was super expensive compared to "normal" SSDs if the QPS numbers got high.

Is that not the case for DBs based on object storage because they cache data before sending it to the object storage? Or because they do some other processing on the DB server before it hits storage?

I don't have any OSS recc, but Google's Stitch works pretty well. Sometimes better than Claude, sometimes worse.

Just looked into how this works. It seems like it's free to integrate for developers, but if you make an app with it, then the users must pay beyond certain limits?

Super interesting idea... but I wonder if they're going to encounter a lot of resistance from users simply because this is far from the mainstream.

Paper looks great. No GitHub link that I can find though. Maybe I'll take a crack at an implementation if I've got some extra free time.

As I said, I agree with you in the case of big outputs. But for small outputs, tool calls can be reliably created from the NL version. There's no need for JSON.

I don't know what they meant by it, and I share your opinion that "AI native" is somewhat meaningless for a programming language like this.

Regarding compilation and static typing, it's extremely helpful to be able to detect issues at compile time when doing agentic programming. That way, you don't run into as many problems at runtime, which of course the agent has more difficulty addressing. Unit tests can help bridge the gap somewhat but not entirely.

What's not stated on their website is that Mojo is likely a bad choice for agentic programming simply because there isn't much Mojo training data yet.