Pinned Tweet
Replying to @sylviapuffs
"it took a RAM shortage crisis for companies to start optimizing shit" THEY CAUSED IT
1
376
5,853
crystl retweeted
Tencent just announced Hy4 Flash on X, then deleted the post 👀
It claimed Claude Haiku 5.5 (high) quality on daily work at 18x lower task cost, priced at $0.07 in / $0.281 out per million tokens. 224B total, 12B active, 1M context.
crystl retweeted
Qwen4 GGUFs will use a new version of Unsloth Dynamic 3.1 for the best performance on Terminal Bench, DeepSWE & more agentic benchmarks.
Today, we're releasing the highest accuracy Qwen3.8-27B GGUFs.
Our new Dynamic V3.1 GGUFs lead by 5–10% on average across Terminal Bench, DeepSWE & more. We’re working on even better version for Qwen4
Run on 12GB RAM.
GGUF: huggingface.co/unsloth/Qwen3…
Blog: unsloth.ai/docs/models/qwen3…
crystl retweeted
Qwen 4 27b coming asap
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery!
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
- $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%.
The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder.
Start building today! 🚀
- Blog: qwen.ai/blog?id=qwen4-max
- API: qwencloud.com/models/qwen4-m…
- Token Plan: qwencloud.com/pricing/token-…
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
crystl retweeted
Well, here we go. Qwen4-Max is officially here
A new architecture, native vision, 1M context, and quite a few improvements we've been working on.
It's been a long time coming, and we're really excited to finally get it into your hands.
As always, give it a try, push it on your own tasks, and let us know what you think. We'd love to see what you build with it!
We know what else you've been waiting for. More Qwen4 models are on the way.
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery!
Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window.
Highlights:
- Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation.
- Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables.
- Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery.
- $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%.
The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder.
Start building today! 🚀
- Blog: qwen.ai/blog?id=qwen4-max
- API: qwencloud.com/models/qwen4-m…
- Token Plan: qwencloud.com/pricing/token-…
More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
crystl retweeted
are these guys new to the fucking internet? /genq
block ip 63.250.52.29
ps: there needs to be a global malicious ip registry and some sort of public ip credit score
this probe thats sending 62k requests to our endpoints is blocked for us now but doubtful that we are the only target.
I should be able to flag their ip as a probe and let other people know
feels like a no-brainer for @Cloudflare
crystl retweeted
Replying to @bilskirnir3124 @thsottiaux
You can't blame LLMs for the incredibly insecure and buggy status quo of large mainstream software, firmware and hardware projects. Linux has hundreds of of preexisting severe vulnerabilities being found and fixed every month. Linux already churned out endless slop without LLMs.
crystl retweeted
Anti-AI activist
-> Rewards AAA games using AI
-> Rewards devs who lie about AI usage
-> Constantly accuse devs who don’t use AI because of “signs”
-> Burn honest indie AI devs on the stake
It’s the modern Salem Witch Trials, ran by people who've never shipped anything.
crystl retweeted
Replying to @bilskirnir3124 @thsottiaux
GrapheneOS code quality has greatly increased due to using frontier models for code review and bug finding. Our port of the Pixel 10 drivers from Linux 6.6 (QPR1) / 6.12 (QPR2) to Linux 6.18 went incredibly smoothly because we were able to find most of the issues in advance.
crystl retweeted
Linux was created as a rejection of robust software architecture and tools. It's a monolithic kernel with no internal isolation written as memory unsafe code. A typo in an obscure driver for unused functionality can easily create a fully privileged code execution vulnerability.
crystl retweeted
Projects can now churn out slop even faster with LLMs but they didn't need them to do it. This was already the status quo for software at large. Being able to so easily find bugs is creating massive pressure to finally use better tools and approaches to create robust software.
crystl retweeted
Show me your github. Let me see your biggest project. I'd like to count the stars. I'll read your issues closed and the PRs opened and closed.
Surely you must maintain a successful highly polished project on your own, all by hand, artisinal code
crystl retweeted
Replying to @thsottiaux
Daybreak Blue lacks cyber access for GPT-6 Astra or GPT-6.1 Sol/Terra. Anthropic has the far more capable Mythos 5.2 (Fable 5.2) for their Defense Access tier.
Daybreak Red lacks an application process. Anthropic has open applications Red Team Access and we recently applied.































