AI News: Opus 5.5, GPT-6 Sol, Jev, Muse and More!

MUMuseMetaUnverified

Meta's easy-setup AI agent, compared to OpenClaw. New: use with AI glasses, voice mode in app and browser, more connectors (Granola), computer use on Mac, and its own email address. Creator uses it for email, calendar and social ideas.

Teased handheld keychain device, compared to a Tamagotchi, for instructing the Muse agent. Still in production, expected for the holidays.

RMRay-Ban Meta AudioMetaUnverified

Ray-Ban Meta glasses without cameras, answering privacy complaints. Creator tested them and said they sound very good.

MRMeta Ray-Ban DisplayMetaLimited in EU

Display glasses with wristband controls getting cycling and transit navigation, better calendar, mirrored notifications, threads, Muse control and hologram calling.

Lightweight 100 g VR glasses with a tethered compute/battery puck, Vision Pro-like passthrough, eye tracking and hand gestures, running Quest content and full-body hologram calling. Coming spring 2027 for about $1,300.

AI platforms on the Meta Horizon engine that turn a prompt into a 2D or 3D mobile game; Create works in a phone app, Studio in the browser.

GLGPT Live 1OpenAIAvailable in EU

API now generally available that lets a voice assistant keep talking while an agent works in the background. Creator built a voice AI news researcher with it, the Agents API and Codex (sponsored segment).

GSGPT-6 SolOpenAIAvailable in EU

Cheaper GPT-6 tier priced at $2 input and $10 output per million tokens, almost as good as Astra at lower cost. In the Mega Bonk test it was decent but below Astra.

COClaude Opus 5.5AnthropicUnverified

New state-of-the-art model, priced at $4 input and $20 output per million tokens. Built a near-faithful Mega Bonk over about 20 hours and produced strong JavaScript and Blender animations. On Artificial Analysis it is the smartest model, but token-hungry, so cost per task stays high.

G4Grok 4.7xAIAvailable in EU

New xAI model, cheaper per token but weaker than the newest models; compared only to last-generation models. Its Mega Bonk result was poor and it fell outside the top 20 on BusyBench.

JEJevType Safe AIUnverified

A decision-making model that outputs structured values (choice, score, null) with a confidence score instead of text. Very cheap (4 cents per million input tokens) and fast; example use cases include sorting files, email triage and moderation. New sign-ups are paused due to demand.

YMYouTube Made On YouTube AI featuresYouTubeLimited in EU

AI custom feeds, Ask YouTube search, live stream auto dubbing, Ask Music, AI thumbnail help, Gemini Omni editing tools, likeness detection, A/B testing of videos and audience-specific thumbnails.

MCMicrosoft Copilot (new)MicrosoftAvailable in EU

Redesigned Copilot with chat, co-work, code and autopilot modes, a choice of GPT and Claude models, and control of Microsoft apps like Excel, Word and PowerPoint. Creator hasn't tested it yet.

Gemini 3.8 Live pairing real-time video generation with speech for face-to-face conversational avatars.

G3Gemini 3.8 Flash TTSGoogleUnverified

Text-to-speech model with over a thousand voices and adjustable voice style, available in Google AI Studio.

GOGemini Omni in Google VidsGoogleAvailable in EU

Omni video model integrated into Google Vids for more dynamic animations, offering a free way to use Omni.

STSpotify Taste ProfileSpotifyUnverified

Feature that shows your inferred tastes and lets you describe changes, using AI to tune your taste profile.

Google plans to launch TPUs into space with SpaceX to test whether hardware survives launch, stays cool and can link satellites.