Bonsai 2 27B, a 27-billion-parameter ternary language model built on the Qwen3.8-27B hybrid-attention architecture, ships at 8.60GB total through the MLX 2-bit pack — a 7.67GB ternary language model p…
MindStudio has no one-click importer for Custom GPTs, so converting a Custom GPT requires a manual rebuild in which each Custom GPT feature maps to a MindStudio block, according to a MindStudio conver…
Agnes AI released Agnes-3.0-Flash Preview, a 33-billion-parameter open-weight multimodal language model under an Apache 2.0 license with a 262,144-token context window and a hybrid attention architect…
A practical architecture for scaling a personal AI second brain into a shared team knowledge base keeps each person's personal agent intact and connects it to a centralized, permissioned database rath…
Google researchers and academic collaborators described Dream RSI, a method that lets an AI system improve its own strategy for guiding scientific discovery by simulating past research history rather …
An MLX-based parallel constrained decoding engine for Apple Silicon evaluates every JSON schema field in a single forward pass, cutting structured extraction latency 5.6x to 7.0x versus standard autor…
An unidentified multimodal AI model called Union Alpha surfaced unannounced on model-routing platforms including OpenRouter with a 262K token context window and no lab claiming it. On a software engin…
Grok bots can be configured to manage an email inbox either by connecting directly to an existing Gmail or Outlook account for auto-labeling, or by giving the bot its own dedicated address through Age…
OnSpace AI has launched a no-code platform that builds, tests, and publishes AI-powered mobile apps from plain-English chat prompts, generating frontend, backend, database, and AI features in a single…
Serena, a tool that connects AI coding agents to a language server through the Model Context Protocol (MCP), gives agents symbol-level code understanding rather than text-guessing, exposing around 21 …
Typesafe's Diogo Almeida is publicly arguing that RLHF has structural flaws and is promoting RLCD (reinforcement learning for calibrated decisions), a training method that rewards outcome accuracy and…
Computer-use AI agents that control desktops via screenshots and commands turn prompt injection into a real-world action risk, since a successful injection can install software, submit forms, delete f…
Google Antigravity's /boost slash command turns a single coding request into a coordinated multi-agent workflow while the user stays on Gemini 3.8 Flash, which landed in Antigravity in early September…
Edge0 released Edge0-35B-A3B-preview, a 35-billion-parameter mixture-of-experts language model built on Qwen3.6-35B-A3B that runs with under 3 GiB of active memory by streaming experts from SSD instea…
An independent research group in Austria has developed two quantization techniques, GSQ and RCO, that compress Z.ai's 320-billion-parameter GLM 5.3 Flash vision-language model from roughly 320GB at fu…
A roughly 400-line markdown skill file lets Claude Code and Codex control a full computer screen through native terminal commands — PowerShell on Windows, AppleScript on Mac, and equivalent shell tool…
A cinematic AI short film built with the text-to-video model Seedance 2.5 and a companion tool called Astra demonstrates sustained character consistency, spoken dialogue, and lip sync across multiple …
Verdant's free mode lets users run its AI coding agent without a subscription or paid credits, with an allowance that refreshes every 5 hours and is also bound by a weekly cap, according to the compan…
A GitHub project called "CUDA for AMD on Windows," built by a developer going by "speed," packages the ZLUDA translation layer with AMD's HIP SDK and ROCm to run unmodified CUDA applications on AMD GP…
Anthropic CEO Dario Amodei published a roughly 9,000-word essay, "We Must Pace the Frontier," arguing AI labs should slow the rate at which they advance capability without halting development, citing …