Qwen3.6 35B A3B vs Claude Sonnet 4.5 🎸 Guitar... Hero, another HTML canvas challenge Went pretty hard on this one ngl, there are a lot of elements and visual effects to stitch together. Best result I could get out of 5 runs on both Qwen set the stage on fire 🤘🔥 ↓ prompt belowshow more

Qwen3.6 35B A3B vs Claude Sonnet 4.5 Made them... fight on the same prompt. 🌸 "Cherry blossom" challenge Both models took exactly 41s to deliver, no retries Oh boy 🤯 Crazier part is that I used Unsloth AI's UD-IQ3_XXS gguf to make it fit in 16GB VRAM...show more

🥊 Qwen3.6 35B A3B vs Qwen3.6 27B Made them... fight on the same prompt 🌊 Ocean Waves, a canvas challenge 35B took 39.7s (at 142tok/s) 27B took 111.6s (at 50tok/s) Both models are really good. The fact that the 35B produced such a strong result in under 40 seconds is seriously impressive. But when you look at 27B's output... it's actually so much better Added clouds, beautiful foam particles and splashing effects, a perfectly straight horizon, more realistic blinking stars, and a nice sun reflection on the water 🏆 Qwen3.6 27B slower but the result slaps Absolutely amazing that both models run smoothly on 16GB VRAM. Used Unsloth AI GGUFs here, Q3_K_S quant to make it fit in the hardware 🔥show more

HOLY MOLY running a 35B model locally on a... MacBook shouldn’t be THIS FAST 🤯 Spent my weekend in atomic.chat testing Qwen 35B vs. Qwen 27B on my local machine. I had them generate a fully animated HTML/Canvas car mini-game (demo below), ... and both models breezed through the physics and parallax scrolling without a hitch! The secret sauce here is the Atomic Chat app. Because it's perfectly optimized for Mac and uses Google's new TurboQuant under the hood, you can run heavy open-source models flawlessly while keeping top-tier output quality 👊 Other perks: → ZERO setup required → Access 1,000+ models completely free → 100% offline and private → Zero API limits ... and MUCH more! I dropped the prompt I used in the 🧵↓ Spin it up locally and let me know what you get!show more

After two rounds of practice, this one went much... faster. A lot of video effects that look complicated become much easier once you break them down into smaller modules. Here are 3 small tips I’ve learned for making this kind of video: 1. Show the final result at the beginning to set expectations 2. Let the decorative elements stay on screen for 2 extra seconds at the start of each segment 3. Cut out the clips where the AI motion goes wrong Full workflow and prompts:show more

New vs Old! Epyc 64-Core CPU vs DEC VAX!... I like to compare weird things on an even playing field, and so I installed NetBSD 10.1 on both a VAX 4000-705A and on an Epyc 8534P, 64-core CPU with 128GB of RAM. The 1993 VAX runs at 112MHz (pretty fast for a VAX!) Then I set them both to building the NetBSD source tree. Epyc on the top, VAX on the bottom. And this VAX is at least 50x as fast as the original VAX 11/780!show more

5 days ago it took 2 GPUs to build... this. today it takes 1. same prompt. same particle simulation. completely different model. Qwen-Coder-Next (80B) on 2x 3090s. 46 tok/s. 564 lines. 2 iterations to get it working. 48GB VRAM across two cards just to hold it. Qwen3.5-35B-A3B on a single 3090. 112 tok/s. 461 lines. first try. cleaner code, fewer lines, better structured. 19.7GB on disk with 4GB VRAM to spare. half the parameters. one GPU instead of two. 2.4x faster. and the output actually improved. this is what happens when architecture catches up to ambition. Gated Delta Networks(Mamba2 variant) hybrid with sparse MoE. 3B active params out of 35B per token. efficiency at the architecture level, not just quantization. the curve isn't flattening. it's steepening.show more

THREE 3090s ON ONE BOARD GIVE YOU 72GB OF... VRAM AND KILL YOUR $200 CLAUDE CODE AND $200 OPENAI BILL people are pulling three used 3090s off ebay for around $2,100 total and stacking them in one tower to build a dedicated ai rig. that pools 72gb of vram for less than what a single rtx 5090 retails for alibaba shipped qwen 3.6 27b in april under apache 2.0. on realworldqa vision it scores 84.1 against claude 4.5 opus at 77.0. on ifbench instructions it lands at 76.5 against claude's 58.0 a single 3090 already runs qwen 3.6 27b with eight gigs of headroom. three of them in parallel handle larger models like deepseek r1 70b and qwen 235b without breaking a sweat a heavy ai user pays $200 claude code, $200 chatgpt pro plus $40 cursor and gemini. that's $5,280 a year and the rig pays itself off before month nine on $8 a month in electricity setup is one shell command for ollama, one to pull the model, one environment variable to point claude code at localhost. cli stays identical, nothing leaves the network, requests stop costing money bookmark this and read the article belowshow more

Dua Lipa and Callum Turner are on an epic... Southern Italy “road trip” 🇮🇹 Worth copying stage by stage! They were recently at “Concettina ai Tre Santi” pizzeria in Naples I went there a couple of months ago One of the best experiences ever Bookmark this place and book ahead!show more

ANTHROPIC 🔥: A new Managed Projects feature has been... spotted in testing on Claude. > Claude takes on tasks and keeps the project organized. > A project is a home for one stream of work. Sessions share memory and instructions so context carries forward, and Claude runs more autonomously. Projects only create and manage cloud sessions. This feature may be built on top of Claude Managed Agents, where each project gets a dedicated cloud environment so Claude can execute periodic tasks and refine project context via "Dreams". It could also be a successor to Conway, which is set to be removed this Friday internally.show more

I went back and did my best to get... it the REAL PS2 Vibez We remember! My love for improving on my craft and yalls input goes a long way. I got a lot of PS3 comments on all my "PS2" videos so I had to go back to work on them. I hope yall like this one better!show more

look what a single consumer GPU just built. gave... Qwen3.5-35B-A3B one prompt: build a cloud GPU marketplace with pricing cards, deploy templates, and a benchmark leaderboard. it planned the layout, wrote the animations, populated the data, and served it. one shot. one HTML file. then i told it to iterate. split the hero, add a floating GPU with neural network animation. glassmorphism on the cards. done. done. done. three rounds, no confusion, no regressions. 4-bit quantized. 19.7 GB. single RTX 3090. full coding agent claude code harness running on localhost. no API calls leaving my machine. no subscription. no rate limits. earlier today i pointed it at my own production website. it curled the HTML, found every broken link, and told me "pretty shell, empty core. would not recommend." then built a better version from scratch. local inference stops being a demo when you actually steer it. the models are there. they understand intent. but you have to meet them halfway with good prompts, clear context, and real project structure. that's the skill gap now. not the models. the steering. more experiments coming. i genuinely cannot stop playing with this thing.show more

this is f**king dangerous someone figured out how to... make Opus 4.8 run on Fable 5's brain with one prompt access to the best model is never guaranteed. It disappeared once already this year. but you can use it forever. here's how: 1. ask Fable 5: "write the operating manual your replacement will run on" (procedures, failure modes, a 5-question self-test) 2. save the output as one .md file and drop it into a new Claude Project as the project instructions 3. switch to Opus 4.8 and now your everyday model runs off the smart one's method, no top-tier price save and bookmark this no matter what full extraction prompt is in the article below: ↓show more

The whole process of putting together the "I hate... cowboys" record and the "go down singin'" record was a long two years of figuring out how do something I've never done before. All the songs were written on guitar and a depth I haven't really hit before. This is the final piece of that puzzle. "The fireside sessions" are both those albums together, all acoustic, featuring the songwriters from both those records. This time in my career ends with both album covers together with me and my dad standing next to each other. It's the first time I've ever been able to see a picture of us as two grown men. I'm moving on to another album early next year, but I hope y'all love these versions of the songs, as they were written. The bluebird last night was a special way to celebrate the ending of these last two years, and the start of something new👏🏼show more

home page hero ✨ Design notes: - "Forever" hero... text dot pixel FX done in unicorn.studio. (I will do a whole tutorial on this later. Unicorn's WebGL engine is absolutely wild and very powerful / robust) - built in Framer - I wanted to recreate the iOS unlock effect where your home screen icons cascade into place in a beautifully timed choreography. This took a lot of careful timing using Framer's "Appear" effect on the hero text and surrounding avatars because it was super important that we didn't lose the legibility of our main message ("Build Your Forever Audience") with all the animations. - If you look closely, the choreography is setup to lead your eye through the hero text first starting with "Build Your" then "Forever" and finally "Audience." - With those text layers in place + the surrounding avatars, there is a slight 1 sec pause before the remaining elements slide in below and above (How it works, CTA buttons, announcement badge, and lastly the main nav). - All told the entire loading sequence is 6 seconds - Custom particle system powers the interactive star field (the stars slowly gravitate to your pointer position, and the star field perspective changes ever so subtly as you move your mouse around on the page) - I have 3 shooting stars made of small white line layers that start out off canvas rotated at different angles that shoot across to another point off canvas at random times on a loop effect. - Given this hero scene is in space, I wanted the surrounding avatar elements to "float" in low gravity mode. For this I used Framer's loop effect that slowly oscillates the layer's y position. I then offset the delay of each element randomly to stagger the floating loop so each avatar floats independently/randomly - The final major treatment for this hero scene was the scroll animations on the avatars. I wanted to create a bit of a warp speed effect when you scroll down, as if the avatars were being pulled or sucked into a worm hole as you scroll down below this hero fold. - To accomplish this, I applied Framer's scroll transform affect set to "section in view" on each of the floating avatars, and set the "scroll to" position of the upper avatars to be much, much further away on the y-axis than the "scroll to" position of the lower avatars. (eg. -1700px on upper most avatars vs. -600px on lowest positioned avatars). This effectively causes the upper avatars to slide up off the hero canvas with much greater velocity than the lower positioned avatars. - And when you scroll back up to the hero section, the inverse happens where the lower positioned avatars "arrive back in place" from up above the hero canvas before the upper avatars come back into the scene and settle in place. - Overall I wanted this hero section to feel alive. The floating avatars, particle system with very subtle star movements, and the Caustics effect on the "Forever" text all sort of move at the pace of slow breathing - which is a great pace to create a sense of life and comfort in your scene. Conversion Results (so far) - When this new Framer site launched along with Calaxy v1.9 release on Base a couple weeks ago, we saw a surge in traffic, around 20k page views in the first few days. - Of those 20k hits, 11k visited the app install page ( - which is our main CTA - We saw around 10K new users in the first week after v1.9 launch Overall I'm very happy with the new site and early performance metrics. Lots of tweaking to do but its a good start. If you are a designer building in Framer - hit me with any questions on the above hero notes. Happy to share more specifics! 👾show more

i couldnt stop thinking about kibum’s gasoline perf yesterday.... he was really pushing to his limits for the best stage performance, trying to overcome his condition but i couldn’t bear this sight :( during the MinKey ment, i was worried key will just pass out. but thank god there’s minho to help him get off the stage 😭 thank you, our shinee’s pillar ㅠㅠ bammie, you did so fucking well but please rest a lot, reenergize, have some nice & healthy foods, get lots of sleep, and come back to us only when you’re fully recovered, okay? 🩵 i love you so much bammie, and im proud of you and of minho. you are both shineeng & devouring on that stage! MIC FUCKING ON ✨show more

"There is absolutely no pressure on Rohit!" — Interim... Coach Sitanshu Kotak on retirement rumors🚨😱🗣️ "See, I don't think a big player like Rohit Sharma can have any sort of pressure. He's too good a player to feel that. Yes, he didn't get runs in the opening two matches, but I don't think that makes any difference. But today, it also looked like he'd probably get a good inning out. But that's okay." "See, honestly, I have played a lot of cricket, and I have seen a lot of cricket. On a day, there are a lot of batters I have seen, they don't get that momentum that they are looking for, and that can happen. You might see a completely different innings from Rohit Sharma at Lord's. So, I wouldn't use the word he was struggling."show more

A CHINESE GUY STOPPED PAYING WEB DESIGNERS $1,800 PER... LANDING PAGE AND BUILT THE SAME KIND OF SITE WITH CLAUDE CODE IN ONE AFTERNOON FOR LESS THAN $70 claude was not just writing code. it was doing the job of a designer, copywriter and frontend dev in one window. in 4 hours, one rough idea turned into the layout, color system, font choices, page sections and the full html/css/js build the real trick was the references. he gave claude 5 screenshots, made it ask 7 questions first, then pushed it toward one clear visual direction instead of accepting another generic ai-looking template the first version already looked solid. the second pass made it feel expensive: better typography, a darker palette, mobile cleanup, cursor effects, 6 micro-interactions and custom hero visuals his old workflow was burning $1,200 on design, $500 on frontend work and another $150 on small fixes every time he needed a new page. now the whole test costs less than $70 and the site still looks like something a $5,000 agency would ship the edge is not “ai builds websites.” the edge is that one person can now brief, critique, polish and launch in one afternoon without waiting 10 days for a designer to send version oneshow more

Grok 4.5 performed GPT Sol level for free! We... gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos Prompts: -robot deathmatch, Tombstone vs Minotaur -a hydraulic press flattening stuff on a conveyor -a semi truck jumping a canyon Outputs: GPT-5.6 Sol: 12.9K tokens, $0.51 (~7 min) Grok 4.5: 10.8K tokens, $0 (~5 min) Muse Spark 1.1: 26.8K tokens, $0.12 (~7.5 min) GLM 5.2: 10.9K tokens, $0.02 (~12 min) Grok 4.5 handled all three scenes genuinely well and got surprisingly close to GPT-5.6 this round. On top of that, it ran on the free tier. GPT-5.6 Sol, the frontier model, put out solid but not standout work. GLM 5.2 rendered all three scenes for pennies, but it came out the roughest of the four. Meta's new Muse Spark burned the most tokens yet still stayed cheap, delivering an average result.show more

Ran some supplies out to Kerr County with Brandon... and met up with Wes Virdell. This is right around the corner from Camp Mystic, where 20 to 25 little girls went missing . Miles and miles of cars, trees, and homes piled on top of one another. First responders told us it’s a huge holiday weekend there with hundreds of out of towners staying on the river bank, which means they really can’t give a solid count on how many people are actually missing. This community is gonna need support in the following days. Keep an eye out on how to help. We’ll try to figure out more tomorrow. Heading back in a couple hours when it’s light outside.show more