AI News: Microsoft Finally Reveals Their Plan!

MMMicrosoft MAI Models (7 new)MicrosoftUnverified

Microsoft AI announced seven in-house models including a flagship thinking model (preferred over Sonnet 4.6), MAI Transcribe 1.5 (state-of-the-art accuracy, 5x faster) and MAI Voice 2 (15 languages). Suleyman says the goal is self-sufficiency and ethically licensed training data.

New coding model compared to Claude Haiku 4.5, claimed more accurate and using fewer tokens.

MIMAI Image 2.5MicrosoftUnverified

Image model with a new ultra-efficient flash variant; now ranked fourth in text-to-image and second in image editing, just behind GPT Image 2.

Always-on personal agent powered by OpenClaw's open-source tech, working at the OS level across Teams, Outlook, OneDrive and SharePoint. Available only to early testers.

GCGitHub Copilot AppGitHubUnverified

Desktop app resembling Codex but letting users pick any model from any provider. The creator couldn't try it because new Copilot sign-ups were paused.

PSProject SolaraMicrosoftUnverified

Open platform for putting AI agents in physical devices, shown with a desk device and a badge with camera and microphone. The creator is unsure who it's for.

MCMayo Clinic Microsoft Healthcare AI ModelMicrosoftUnverified

Mayo Clinic and Microsoft are collaborating on a frontier AI model for healthcare. Suleyman predicts medical superintelligence within two to three years.

NRNVIDIA RTX SparkNVIDIAUnverified

Combined CPU/GPU chip with up to 128 GB unified memory for running large local LLMs, announced at Computex. Creator saw demos; likely expensive and not yet available.

NRNVIDIA RTX SparkNVIDIAUnverified

Surface laptop built on RTX Spark; creator saw impressive graphics and local LLM demos. Price and date unannounced, expected to be expensive.

N3Nemotron 3 UltraNVIDIAAvailable in EU

550B-parameter open-weight model, state of the art among open models for agent productivity and cost-efficient.

G4Gemma 4 12BGoogleAvailable in EU

Open-weight model that runs on a laptop and nearly matches Gemma 4 26B on benchmarks.

MMMiniMax M3MiniMaxAvailable in EU

Coding model with 1M token context, claims to beat GPT 5.5 and Gemini 3.1 on SWE-Bench Pro, and is inexpensive.

CWCodex Windows Computer Use & Remote ControlOpenAIAvailable in EU

Codex app now supports computer use on Windows and remote control of Windows devices from ChatGPT on iOS or Android.

CPCodex Plugins and SitesOpenAILimited in EU

Codex gains role-based plugins (data analytics, sales, product design, investment banking, etc.), annotations, and a Sites feature that gives shareable URLs for apps and websites.

CDChatGPT DreamingOpenAILimited in EU

Updated ChatGPT memory with a new dreaming feature that better saves and recalls past conversations. Creator admits not fully understanding the changes.

HAHermes AgentNous ResearchAvailable in EU

Nous Research released a desktop app for tracking Hermes agents, with a Codex-like design. Creator hasn't tried it yet.

I4Ideogram 4.0IdeogramUnverified

Open-weight image model with API, ranked ninth on arena.ai (highest open model), good text rendering, native transparent backgrounds, and trained with bounding-box region descriptions.

O2OpenClaw 2.0OpenClawAvailable in EU

New image model ranked number two on arena.ai, beating Nano Banana in text-to-image, behind only GPT Image 2 in most categories.

K2Krea 2 TurboKreaAvailable in EU

Faster version of Krea 2 that generates images in about 2 seconds with similar quality.

GIGrok Imagine 1.5xAIUnverified

New video model with dialogue; the sample still has an obvious AI lip-sync quality.

RARunway Aleph 2.0RunwayUnverified

Video-to-video editing model; creator tested turning a man on stage into a humanoid robot and found the result okay but would stick with Gemini Omni.

Open-source voice model claimed to be the most emotive in the world; samples sounded like real human recordings.