Waifus incoming, GPT 6 Sol, Grok 4.7, Opus 5.5, Mimo 2.6, Step 5, OpenMuse: AI NEWS
AI Search showed 24 AI tools in this video on . 9 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-39.
Tools shown in this video, with timestamps
Interactive world model generating explorable 3D worlds from text or an image, with implicit 3D-aware memory via a reconstructed point cloud for consistency. Base and distilled models released, about 147 GB total.
Open 6B text-to-image model for visual design (UI, infographics, posters) with 2K output and alpha transparency; a companion Design Layer model splits flat designs into layers. Ranked first among open weights on Artificial Analysis UI/UX leaderboard, MIT license.
Reconstructs a single image into a 3D scene with separate objects by detecting objects, estimating depth and generating geometry. Usable in Blender or physics simulators; code and dataset released, about 6.1 GB.
7B world action model that takes goals, camera views and joint positions, predicts actions and resulting world state. Beats Cosmos 3 Nano with fewer parameters and faster speed; weights released (14 GB).
Text-to-speech models supporting voice design from text prompts, expressive meta tags and 100+ languages. State-of-the-art on TTS benchmarks, cheap, available via API, AI Studio and Gemini Notebook.
Live demoOpen-weight frontier model; the 1T Pro is the leading open model on the Artificial Analysis index and very cheap per task, with a 311B Flash variant. Also released Mimo Desktop harness.
Multimodal agent MoE model with 600B total and 27B active parameters and 1M context. One point behind GLM 5.3 on Artificial Analysis but about three times cheaper; open release planned for October 15.
Generates consistent 3D worlds with video, depth map, camera trajectory and 3D reconstruction simultaneously. Weights under 12 GB plus training and eval scripts released.
Open-source MIT-licensed personal agent with browser, terminal and files, connecting to Gmail and Calendar; works with any model. Alternative to Meta's Muse.
Open-source system-one model built on Qwen3 8B using contrastive learning to score actions against situations; nine times faster than Jev while matching performance, shown playing Dino Run.
xAI's latest model; behind frontier models on Artificial Analysis intelligence and tied with Mimo 2.6 Pro, and per the creator not cost-efficient. Available in Cursor, Grok Build and API.
Mid and light tiers of the GPT-6 family (Astra top, Sol mid, Luna cheapest). Sol Max is a few points below Astra but over three times cheaper; Luna is very cheap. Available in GPT work and code on paid plans, Luna in desktop app for free users.
Anthropic's latest model, strong at 3D, design, coding and navigating interfaces, but not significantly better than GPT-6 Astra and much more expensive per task. Creator prefers GPT-6.
About 950 Claude agents using 210M tokens over 21 hours searched DNA for unusual reverse transcriptases and flagged a possible CRISPR-related system. Function is unknown and biotech critics say significance is overblown.
Transformable robot switching between quadruped and bipedal humanoid modes, priced around 20,000 yuan (~$2,800), shipping from October.
Live demoSuper-realistic humanoid robots with realistic faces and hair; first batch now being delivered.
Human-sized dexterous robot hand with 22 degrees of freedom and impact torque protection, from $6,500. Demos include piano and scissors.
Live demoHumanoid robot model S1 trained via self-play reinforcement learning in Isaac Sim over 140+ years of soccer with an automatic training ladder, then deployed to real robots.
Live demoOpen MoE model with 208B total and 16B active parameters; performs close to much larger models and leads open models on ARC-AGI-2. Full model 577 GB, FP8 300 GB.
Multimodal embedding models mapping text, images, video and documents (Omni 3B also audio) into one space; state-of-the-art among similar sizes, Apache 2 license.
1B parameter model specialized for hard competitive math, scoring about 94% and beating models far larger. Around 2 GB total, code and models released.
Open-weight cybersecurity model fine-tuned from GLM 5.3 and compressed from 1.5 TB to 328 GB; found 23 of 32 vulnerabilities versus 25 for the full model.
Google's plan for space-based AI data centers; Trillium TPUs survived vibration and radiation tests, with a two-satellite prototype test planned next year.
Tiny 100M-parameter image model (417 MB) that can run on a phone; image quality is low but it is about 60 times smaller than Z Image.
