Cybersecurity chaos, AI for DNA, Kimi K3.1, Claude Duet, Dessa, Fibo 2, Minimax 3.1: AI NEWS
AI Search showed 21 AI tools in this video on . 11 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-43.
Tools shown in this video, with timestamps
Vision language model built on Qwen3 VL with a component for pixel-level object identification. Handles detection, counting, pointing, text recognition, segmentation and alpha matting in one model. Released open under Apache 2 in 4B and 2B variants.
Tiny speech-to-text transcription model family (163 MB smallest, plus and pro versions up to 1.25 GB). Creator tested it in a Hugging Face space. Competes with Whisper Large v3 and Parakeet; license allows commercial use with attribution.
Live demoOpen image editing model using natural language, with multi-reference support. Creator tested outfit transfer, upscaling, restoration, character sheets and lighting correction. Ranks first on the Artificial Analysis editing leaderboard, slightly above Qwen Image 2.1, but its license is non-commercial.
Live demo14B open model for DNA understanding and generation, started from Qwen 3 and trained on 300B DNA tokens plus RL. Reportedly beats larger frontier models on DNA benchmarks. Released under Apache 2.
Real-time interactive world model that generates short segments with memory for consistency. Outputs 720p at 24 fps. Code, training scripts and models are released under Apache 2.
Hi3D open-sourced its version 1 model for detailed 3D asset generation from prompts or images. The latest version 3 remains closed. Total size is about 7.74 GB and it compares favorably with other open 3D generators.
3D reconstruction model that builds 3D scenes from video or images. Existing models were improved using unlabeled ordinary videos converted to depth and motion cues. Open source with training scripts.
Open text-to-speech model with natural language voice control, voice cloning, emotion and sound-effect tags, and multilingual plus Chinese dialect support. Two sizes (about 3 GB and 18 GB, the latter with multi-speaker podcasts), MIT license. Slightly ahead of Fish Audio's model on Artificial Analysis.
Live demoOpenAI blog showing thousands of GPT-powered agents analyzing a mouse brain gene perturbation dataset in 32 minutes. The agent team cited over 700 references versus 11 for a single agent. It surfaced hypotheses about cholesterol regulation and Alzheimer's/Parkinson's.
World model that turns text or images into explorable 3D environments with characters and animations, kept consistent via 3D matter representation. It gets a higher win rate than Meshy and Tripo. Not open source but free to try on their platform.
Claimed first world model for engineering. It breaks mechanical objects into 3D parts with CAD sketches and steps, and generates architectural floor plans. Faster and cheaper than frontier models on its tasks; waitlist only.
3 trillion parameter model with a 1M context window, roughly on par with frontier models on benchmarks and cheaper per task. Available on Kimi's platform and API, with open weights planned for release later.
Model specialized in generating editable vector graphics such as logos and illustrations, and converting existing images into vectors while keeping detail. Available in the Ideogram app.
Real-time speech-to-speech voice model that learned from human conversations without text, picking up accents, breathing and humor. Access is limited to a small group of AI safety researchers.
Live demoProactive assistant agent that monitors work, summarizes emails and calendar, and completes tasks even when the computer is off. It integrates with 6,000+ tools and runs on Claude Opus 5.5 with its own virtual computer. Waitlist only, US Pro/Max plans.
Humanoid robot positioned as a heavyweight service and guard robot, shown wrecking a car in a demo. Reportedly has about 8 hours of standing endurance; few specs known.
Open 35B mixture-of-experts model based on Qwen 3.6, optimized for scientific research with multi-agent evidence gathering and cited reports. Beats similar-sized models on benchmarks; weights and code released (full ~72 GB, Q4 GGUF 22 GB).
Fast, affordable upgrade of the stealth model Space Bunny seen on OpenRouter. Competitive with leading open models on agentic coding and automation, and very cost-efficient. Weights to come later; available via API now.
Open model that stores information across conversations in a compact internal memory state instead of rereading chat history. Comes in a Nano version (3B active) and a Flash version (288B total, 17B active). Strong on persistent memory benchmarks.
Open medical MoE model with 124B total and 5.1B active parameters, specialized from Ling 3 Flash. Competitive on clinical knowledge and medical reasoning benchmarks; full model is about 255 GB. The creator attributes it to Anthropic, which appears to be a misstatement.
Acceleration method for an open reference-to-video model, cutting 50 steps to 8 with up to 12x lower latency on GB200s. Released variants include ComfyUI, FP8, FP4 and Apple silicon.
