Thinking Machines released Inkling 2 Mini, an open-weights 26.4B-parameter Mixture-of-Experts model designed for fast local agentic work.
Keep up with software engineering, without the endless scroll.
Follow the communities developers read and the engineering teams that build the tools. Snapbyte.dev turns them into one summarized digest on your schedule.
Inkling 2 Mini fits in 20GB as Frostbite hits 4.59 times the performance per watt
AI & Machine Learning
Thinking Machines Lab released Inkling 2 Mini on Nov 20, 2026, an open-weight multimodal 26.4B-parameter MoE model with 3.4B nominal active parameters (5.4B effective per token) under Apache 2.0.
An open 27B language model can rewrite its own weights through reasoning and forward passes, without gradients or training, improving itself and repairing a damaged copy.
From the places developers already read
How it works
Pick what you follow
Choose the communities and engineering blogs, the topics, and the hour your digest arrives.
We read everything
Every story is matched against your topics and ranked by how the communities received it.
You read the best of it
The top stories arrive summarized, with the discussion and a link to each source.
Ranking follows your choices, so two people reading the same sources get different digests. How ranking works
Topics
Software engineering by topic
Browse the latest stories on any topic, then turn the ones you care about into a digest.
Sources
Communities and engineering blogs
Community sites bring the discussion. Engineering blogs bring the work itself, from the teams that build the tools.
See every sourceCompare
How it compares
See where a scheduled digest fits beside newsletters, feeds and community sites.
One digest, on your schedule.
Summaries in English, Spanish, French, Italian, German and Turkish, delivered by email at the hour you choose.