2coffee.dev

Everyone is buzzing about the news that GPT‑6 Astra has officially been launched by OpenAI.

Astra scored 99,9% on ARC-AGI-3 and 100% on ExploitBench, 98% on FrontierMath Tier 4, and has helped solve many long-standing problems in mathematics. In short, Astra has surpassed the basic human action effectiveness level at 96%, and overall is no different from an ordinary person. Sounds familiar, right, that's right, this is the legendary AGI (Artificial General Intelligence), everyone.

Many people believe that Astra's 99.9% score was actually achieved thanks to OpenAI using its own specially optimized framework (responses API harness). When run with the test organizers' neutral framework, Astra only scores around 62% - 62.7%. Astra's Intelligence Index on Artificial Analysis is only 61 points, ranking alongside Grok 4.6 or GPT-5.6 Sol and even behind Opus 5 🫩. In general, scores only reflect part of the picture and shouldn't be trusted absolutely based on advertising like that.

Once again the information above may make some people feel pressured and that it's pointless to have to constantly learn new frameworks or tools, because they risk becoming obsolete after just a few months. Many programmers are deeply concerned about their career prospects amid the rapid-fire release of new models that are becoming increasingly refined. Not to mention if they fall into the hands of "bad actors" then what will happen to humanity 😩

Huh, I looked away and back and thought only Codex was down. Turns out they all went down together, Claude, xAI are gone too, guys.

Gemini is still alive and kicking, and the providers in China don't seem to be affected either. I wonder why they all went down together? Probably AWS or Cloudflare 🥶

status.openai.com status.claude.com status.x.ai

NVIDIA to acquires Hugging Face.

One side is a chip manufacturer, the other is a platform hosting millions of AI models. Wow, what products are we going to get next 😱

I haven't seen anyone mention it at all, so I forgot too. A few weeks ago Meta launched the Agents Code tool Muse Code, along with the Muse Spark model families made specifically for it. I tried Muse Spark 1.2 but wasn't all that impressed, probably because I haven't had it tackle my everyday work yet so I haven't seen much value.

Last weekend Muse Code shed its experimental (beta) label to move into a stable release, today Meta followed up by releasing Muse Spark 1.3 with extremely impressive scores, far surpassing its previous 1.2 predecessor.

Well I guess I'll have to give it a try and see 😁. Muse Code is selling monthly renewal plans now, starting at just 5$ and apparently you get 10-50 requests every 5h 🫩

Early in the morning I read the news that Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, adding to their 2 most premium models then in the evening I read more news that Google released Introducing Gemini 3.8 Flash and 3.8 Flash Cyber - 2 more models in the Flash line - meaning a balance between performance and price.

It seems like Google is continuously releasing models that ease the financial burden on users instead of focusing on the premium segment like the Pro lines as they did before or what 🤔? Or maybe they're also keeping something absolutely massive under wraps so that later they can suddenly go... surprised? 😅

So here's the thing, I have a habit of fighting procrastination by trying to... put things right in front of my eyes or keep them within reach.

Sometimes there are lots of things I need to do, want to do but a lot of the time they get interrupted by the thought that there's something else more important that needs to be prioritized first. Then day after day it stays the same, the thing I intended to do still hasn't gotten done but in my head I'm still convinced I'll do it, the only question is when?

So I just have to leave it where it'll hit me right in the eyes, or more extremely make it get in the way of other work before I might finally get around to it. I've been too lazy to read books these past few days, even though I've already left a whole stack beside me that I'm planning to read, it's just that where I put it is too neat and it accidentally becomes hidden in plain sight. The other day I put a book right in front of me. Between the keyboard and my 2 arms resting on the desk to type, so the book accidentally became something that was in the way but I absolutely refused to move it. So it just sat stubbornly in one spot, every now and then when I looked down it was like seeing a reminder saying "pick me up and read me". And yet it really worked 😆

People say that JavaScript/TypeScript is actually the "native" language of LLMs because it is the richest source of data used to train large language models. Recently, Google published an article titled Why Go is an Ideal Language for AI-Assisted Software Engineering to explain why Go is actually the ideal language for collaborating with artificial intelligence.

It's easy to understand, too, because Go has a simple design philosophy, high performance, can do pretty much anything, and is also very approachable. What do you think about the reasons they gave above? Personally, I think Go is well worth using, but I haven't tried combining it with AI yet to see whether it really lives up to the hype 😁

I suddenly realized that there hasn't been any post mentioning Qwen3.8 27B yet - this is the next small-to-medium-sized model released by Alibaba as they promised, following the success of Qwen3.5.

If you didn't know, people online have been talking about how Qwen3.5 is a relatively good model for running locally for programming, balancing performance and speed, and being capable enough for some basic programming tasks. This 3.8 version continues to prove that, although the number of parameters isn't as varied as before, it makes up for that with a reasonable size that allows it to run on many devices users own.

I haven't had many opportunities to try out this version yet, but it seems to be getting quite a lot of attention!

Reading this article, I find it pretty thought-provoking, everyone: AI is removing the middle class of software engineering.

The author argues that artificial intelligence has now removed the speed limit on software development. I don't know about you, but I really do feel that's the case, because these days, if you ask me to sit down and type out code line by line, it's honestly... very difficult. Instead, I find myself doing more of the work of a problem solver, reviewing and evaluating things. This is both beneficial and harmful. The benefit is speed, while the downside is feeling like I'm able to accomplish a lot, when in reality that may not necessarily be the case.

Large language models, AI Agents/Harness... are now changing the hierarchy of people working in the field. They help new programmers work dramatically faster, while those at the senior level are able to fully leverage their power based on the years of experience they've accumulated. So what do you think will happen to the people in between those 2 levels?

Finally managed to carve out some extra time to revamp the website's homepage 🥳

Did you spot what I was going for 🤓? What do you think of it?