5/ They are open-sourcing the Lite series in two sizes, an 8B dense model and an A3B mixture-of-experts version. The product has some exciting new updates. An 8-step distilled LoRA now open-sourced: inference cut from 23s to 2s on H100 (100 NFE -> 8 NFE).
ComfyUI is now
@kimmonismus
-
Open-sourcing Lite series with 8B and MoE models plus fast distilled LoRA
By
–
-
AI generates 1950s comic cover with precise spatial relationships
By
–
4/ I tried a completely different style for this 1950s comic cover. The model understands spatial relationships and layouts well enough to place characters and text exactly where they belong. It makes generating posters or comics in a single flow much easier.
-
High-density info rendering with structured icons and clean text
By
–
3/ It also handles high-density info rendering very well. I generated this awareness infographic to see how it manages complex layouts. It kept the text clean and the icons structured, which is a common pain point for open-source models.
-
Composer 2.5 scores 62, price difference makes ‘slightly better’ not worth 60x
By
–
Composer 2.5 scores 62 on the Artificial Analysis Coding Agent Index. The two models above it score 65 and 66. The price difference: $0.07 per task vs. $4–5. At some point "slightly better" stops being worth "60x more expensive," and most engineering teams crossed that point a
-
Local-first automation: Scalable token for non-technical users, no cloud risk.
By
–
4/ The main reason I keep this running is that it is local-first. Everything runs on my own device, so there is no cloud dependency or data leakage risk. For anyone non-technical who wants to automate tasks without learning how to use a Terminal, it completely scales your token
-
AI agent builds zero-coding tools to automate repetitive tasks
By
–
3/ Having an AI agent build functional tools with zero coding required completely changes how repetitive work gets done. It pulls multi-source feeds, strips out layout clutter, and refreshes instantly, saving me from manually opening and jumping between 15 different browser tabs
-
Building a news dashboard with KroWork’s thought process
By
–
1/ I finally tried building my own tool with KroWork to see if this "Chat-to-software" thing is as easy as they say.
— Chubby♨️ (@kimmonismus) 21 mai 2026
I asked the agent, Kro, for a news dashboard, and updating to the latest version lets you see the full KroWork thought process mapping out the system logic. It… pic.twitter.com/vC6LPMym2D1/ I finally tried building my own tool with KroWork to see if this "Chat-to-software" thing is as easy as they say. I asked the agent, Kro, for a news dashboard, and updating to the latest version lets you see the full KroWork thought process mapping out the system logic. It
-
Qwen 3.7 Max achieves 10x speedup in autonomous 35-hour run
By
–
Alibaba released Qwen 3.7 max. Benchmarks incredible. Their new model ran autonomously for 35 hours, made 1,158 tool calls, and achieved a 10x speedup – on a single attention kernel. This isn't "AI improving itself across the board." It's a model grinding through
-
Spiegel article asks if Nobel laureate’s AI use is scandal
By
–
Der Spiegel, arguably Germany's best-known news outlet, has published an article that couldn't be more emblematic of the mood in Germany. Translated, it reads: "The Nobel laureate in Literature admits to using AI for her books. Is that a scandal?" Let that sink in for a second,
-
Anthropic pays SpaceX $1.25 billion monthly for compute, majority revenue
By
–
Anthropic is paying SpaceX $1.25 billion per month for compute. Per month. That's $15 billion a year flowing to a company whose total annual revenue is $18 billion. One AI lab is about to account for the majority of SpaceX's commercial income. We only know this because SpaceX