


growth & ops


clairic is redesigned and now open-weights! we went back to the drawing board and overhauled the architecture for it to excel at extraction. new: - clairic (27B) and clairic-flash (9B) - image extraction greatly improved - PDF support - weights release! blog.cloudflare.com/open-cla…


Really sad to see this today, this is unreasonable interference that fundamentally harms the free and open Internet
Quad9's forced withdrawal from France highlights the dangers of mandating DNS-level blocking. Read more: cfl.re/4f0cowv


An Advent calendar, but each day a part of our infrastructure fails.


everything’s a new algorithm now. score centering? new algorithm. FP4 scaling trick? new algorithm. RL env for regional tax law? new algorithm. pretraining on last gen’s vibecoded repos? new algorithm. prompting model to write better docs? new algorithm. it’s getting exponential


Introducing e-means, an algorithm beating k-means in 10x fewer epochs. With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵

Qwen announces smaller Qwen4 models will be released soon! 🔥 Qwen4-27B will run on 16GB RAM setups.
📢 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. 👀

Has anyone tried training models with the power of friendship? Maybe a model trained by people that love each other would solve alignment.

Inkling 2 Small took a while... but it's more interesting than I initially thought! Using attention residuals (like Kimi K3), but without the linear-attention hybrid. Food for discussion for the end-of-the-year architecture trends analysis!



There is something very elegant about the way they train their models
Training with fewer bits is faster, but quantization error can interfere with learning. Today on Connectionism, we detail how we trained Inkling 2 Small with a native 4-bit forward pass from the first step of pretraining through the final RL rollout. thinkingmachines.ai/blog/end…