#Research
Curated coverage on «Research»: releases, research, benchmarks, open-source.
NEWX: @gengdaJLinger Runs an AI Agent on a 16 GB Mac With No Token Fees and Promises a 5x Speedup
X: @NVIDIARoboticsNVIDIA Releases 100,000 Hours of Human Demonstrations to Train Robots
TG: @bezsmuzi (Русский ИТ бизнес)US AI Got 1.91x Better, but Costs Rose 6.19x: Return per Dollar Fell 69%
X: @cheese_cakee_9Inkling 2 Mini packs a million-token context into 16 GiB instead of 96
X: @simplifyinAIMicrosoft Ships a 100B-Parameter Model Running on a Single CPU: Responses Up to 6.17x Faster
X: @grokMicrosoft's bitnet.cpp speeds up LLMs on ordinary CPUs by up to 6.17x and cuts energy use by 82%
Habr AIEvoNova Labs Releases a Chaos Benchmark of 10 Failure Modes for Always-On AI Agents
Habr AIHow to Test Whether Alisa AI Is Flattering You When Choosing SaaS: A Three-Phrase Test
Microsoft Ships a 100B-Parameter Model Running on a Single CPU: Responses Up to 6.17x Faster
Microsoft has open-sourced bitnet.cpp, which runs large language models on an ordinary CPU, generates responses up to 6.17x faster, and uses up to 82.2% less energy.

Microsoft's bitnet.cpp speeds up LLMs on ordinary CPUs by up to 6.17x and cuts energy use by 82%
A fact-check of a viral claim: bitnet.cpp does speed up 1.58-bit-weight models on ordinary CPUs by up to 6.17x, but no trained 100-billion-parameter model exists…

EvoNova Labs Releases a Chaos Benchmark of 10 Failure Modes for Always-On AI Agents
An agent that aces tests can still break after four hours in production. EvoNova Labs has published an open benchmark of 10 common failures and a set of defensive…

Tencent Releases Hy4 Weights: 49 Billion of 770 Billion Parameters Active per Token

Five Free GitHub Repos Open 1,400+ AI Models Behind One Endpoint

Thinking Machines Releases Inkling 2 Mini: A Laptop Model Six Times Cheaper Than Qwen3.8 27B

Four Free Programs Generate AI Video on a Home PC with 8 GB of VRAM

Imagecaster trains a neural network on your face from 10–100 photos and tries on clothes at home
The open-source Imagecaster runs on a home GPU, trains a neural network to draw a specific person from 10–100 photos, swaps the face, and tries on clothes from…

Microsoft's Agent Hand-Wrote Neural Network Weights and Hit 100% Accuracy on 13 Tasks
Microsoft's coding agent wrote neural network weights in Python with no training or gradient descent, reaching 100% accuracy on 13 tasks with median models of 166…

Author of mod_wsgi Uses AI to Write Muru, a New Python Application Server in Rust

Inworld Open-Sources Voice Isolator: A Model That Cleans Speech of Noise in Real Time

Android work profile closes real-IP leak that Happ and v2rayNG don't catch

Anthropic and Claude Claim Proof of the Riemann Hypothesis, Stanford Responds Without AI

A designer with no coding experience spent 3 days and 700 revisions getting his AI-built video clipping extension to work

Anthropic mathematician says Claude helped prove the Riemann hypothesis in six weeks

North Korea Shows First Chatbot "Source of Wisdom": Science Only, No Politics

Instead of Cursor: Six Open-Source Coding Agents, OpenCode Has 228K Stars on GitHub
Six free coding agents are worth trying before you commit to a Cursor subscription. The most popular is OpenCode, with 228K stars on GitHub, a sign of developer…

Open-Weight Inkling 2 Mini Matches Qwen3.8-27B on Independent Benchmark, Runs on 16 GB of Memory
Thinking Machines has released an open-weight model with 26.4B parameters that reuses some layers three times and can run locally on 16 GB of memory.






