LunalFresh

Curating the cool | Stay Fresh 🌙

[LunalFresh @lunalfresh]

@ultrathinktrash @thsottiaux https://t.co/RnafT47p6E

[Tibo @thsottiaux]

Hi.

It is done.

[LunalFresh @lunalfresh]

@thsottiaux Thanks for the reset as always. Hope you and the rest of the OpenAI team all have a great week too.

[LunalFresh @lunalfresh]

As part of that I will also be releasing a couple ChatGPT sites with the first being released today to show speedups per commit. It’s aimed at clarifying baselines and just being easier to digest and comprehend while showing the real improvements.

https://t.co/OxXWfk7lcs

[LunalFresh @lunalfresh]

https://t.co/TWrGL7rAxQ

[LunalFresh @lunalfresh]

Ever since I started looking at my options for local models for my 9070 XT the options are not great to say the least. Nothing really feels like it’s meant for the 9070 XT. I’d like that to change and that’s why I submitted my first non-personal PR to llama.cpp.

[LunalFresh @lunalfresh]

Huge fan of the work @0xSero does. He was the one to inspire me to me start researching and looking into local AI. Everyone deserves to access AI and it makes me happy to see people putting in the work.

LETS MAKE AI LOCAL

https://t.co/mG5vxyIAHJ

[OpenAI @OpenAI]

We’re making better intelligence easier to access in ChatGPT for everyone:

- GPT-5.6 Sol now powers both Instant and deep reasoning for Plus & Pro users, delivering more factual, focused responses.

- Free & Go users get unlimited text chats with GPT-5.6 Luna starting tomorrow.

[LunalFresh @lunalfresh]

https://t.co/EAT7OgHYRn

[LunalFresh @lunalfresh]

Actually as it turns out there’s an update for free and go users today- they now get access to GPT-5.6 Luna with thinking via a new thinking toggle. There is a new toggle as you see below for plus and pro users as well. Glad to see this- AI should be for all. Not just the rich few.

https://t.co/YVry3ZevNf

[LunalFresh @lunalfresh]

Did anyone else get this new toggle in ChatGPT on iOS? I love when @OpenAI does things like this. Genuinely reminds me of Wii era of gaming. Always something to find with a neat sound or cool vibration to see/feel when moving through the menus.

[LunalFresh @lunalfresh]

Did anyone else get this new toggle in ChatGPT on iOS? I love when @OpenAI does things like this. Genuinely reminds me of Wii era of gaming. Always something to find with a neat sound or cool vibration to see/feel when moving through the menus.

[LunalFresh @lunalfresh]

This is why projects from @PrismML, @Eschalabs, and @syzygyeng interest me so much. They seem to be among the few who treat intelligence per GB and practical consumer inference as the primary goal- not just an afterthought. I’d love to see this become a much larger focus.

[LunalFresh @lunalfresh]

I love open source and local AI, but I think there’s a real disconnect in the community. We aren’t prioritizing models designed to be as intelligent, compact, and efficient as possible. 9/10 times models are designed for datacenters and then compressed until they barely fit on consumer hardware.i would love to see more models designed from the ground up to run well on a NORMAL PC.

[LunalFresh @lunalfresh]

I love open source and local AI, but I think there’s a real disconnect in the community. We aren’t prioritizing models designed to be as intelligent, compact, and efficient as possible. 9/10 times models are designed for datacenters and then compressed until they barely fit on consumer hardware.i would love to see more models designed from the ground up to run well on a NORMAL PC.

[LunalFresh @lunalfresh]

@Pokee_AI Definitely looking forward to the future release. Where can I try out the model currently?

[LunalFresh @lunalfresh]

@syzygyeng Would love to see this done with deepseek v4 flash 0731. Continue the great work consumers need more models like this.

[LunalFresh @lunalfresh]

@sama 750 tk/s faster maybe? 👀

[LunalFresh @lunalfresh]

@DataSculpting @huggingface Are you using xet or http? I was having these instances when using xet where it would be 5-10 minutes if not more sometimes just sitting idly. I decided to try http and it’s slower but it doesn’t get stuck like that. It’s worth a shot if you hadn’t tried that yet.

[LunalFresh @lunalfresh]

Are we being serious, @huggingface? I reported a LoRA made to generate child-looking characters and your system opened a PUBLIC DISCUSSION with my account attached. Then a mod replies “what exactly is illegal or restricted here?” BRO. THAT IS WHY I REPORTED IT. REVIEW IT.

[LunalFresh @lunalfresh]

@dkundel

[LunalFresh @lunalfresh]

I have not found goals to be worth using at all. They pretty much never accomplish the goals I provide them no matter how descriptive or sparse I am in my prompt. I even tried following a guide and having gpt-5.6 Sol proofread and double check my prompts and it still gives me the same mediocre results.