46 / Nov 10 – Nov 16, 2025 · 23 RELEASES

AI News Week 46, 2025

From Gemini upgrades to RAG science, this set focuses on production-ready enterprise AI.

Language Models (3)

VibeThinker 1.5B
WeiboAI · Reasoning
Diversity-driven optimization turns a 1.5B-parameter lightweight model into a strong logical reasoner rivaling larger LLMs.
GPT-5.1
OpenAI · Text Generation
Latest flagship release focuses on multi-turn orchestration, lower latency tool use, and safer agentic execution.
ERNIE (Baidu)
Baidu · Multilingual
Enterprise-ready ERNIE suite highlights multilingual models, search-native workflows, and creative tools for China-first deployments.

RefVTON / EVTAR
360CVGroup · Image Editing
End-to-end virtual try-on pipeline uses unpaired visual references to deliver realistic garment transfer.

Time-to-Move
Tsinghua / PKU · Text-to-Video
Training-free dual-clock denoising controls object and camera motion for text-to-video without retraining base models.

Step-Audio-EditX
StepFun · Text-to-Speech
Open-source RL-based audio editor delivers expressive emotion/stylistic control plus zero-shot TTS in one stack.

Marble: A Multimodal World Model
World Labs · 3D Generation
Turns text, images, or video into editable 3D scenes with export workflows for spatial design pipelines.
PhysWorld Web Demo
PointsCoder · 3D Generation
Physics-grounded world generation demo shows controllable trajectories and interactive scene stitching.

Lumine
MiHoYo Research · Agent Framework
Shares recipe for training generalist agents capable of fighting, puzzle solving, and UI manipulation inside Genshin-scale worlds.
SIMA 2
Google DeepMind · Embodied AI
Gemini-powered embodied agent that reasons over long-horizon tasks and multimodal prompts in virtual 3D environments.
Coinbean
coinbean · Agent Orchestration
Open-source AI native research copilot for crypto market data with multi-agent workflows.
Automaze CCPM
Automaze · Agent Orchestration
Customer Conversation Process Manager delivers drag-and-drop orchestration for multi-agent customer ops.

AI Engineering Toolkit
Sumanth Raghav · Evaluation & Benchmarks
Living repository cataloging agent frameworks, evaluation harnesses, orchestration recipes, and infra checklists.
Prompting Tips for Bolt
Bolt.new · Prompt Engineering
Hands-on prompt engineering tips for shipping multi-agent experiences inside Bolt’s AI-native editor.
MLSys Book
mlsysbook.ai · AI Infrastructure
Open textbook covering scheduling, serving, and observability for modern ML systems engineering.
Harvard CS249r (Edge AI) Book
Harvard Edge Computing Lab · Edge Deployment
Coursebook dives into on-device ML stacks, optimization, and deployment case studies for constrained hardware.
Unipile Pricing API
Unipile · Web Scraping
Unified API for scraping competitor pricing pages and omni-channel communications for go-to-market teams.
Anything LLM
Mintplex Labs · Data Integration
Self-hosted knowledge workspace supporting vector stores, connectors, and provider-agnostic chat.
OpenPipe ART
OpenPipe · Prompt Engineering
Adaptive Response Tuning helps teams reproduce API-only models locally with small fine-tunes.

Teaching AI to See the World More Like Humans Do
Google DeepMind · Research Paper
Multi-step alignment pipeline helps vision-language models mimic human “odd one out” intuition and boosts reliability.
Startup Funding Tracker
FoundersDay · Survey & Overview
Live tracker catalogues this week’s AI startup raises with filters for vertical, geography, and round size.
Context-Rot Research Note
Chroma · Research Paper
Explores how rotating context windows impacts retrieval-grounded chat quality and offers mitigation tactics.
Task Arena
Dimension · Benchmark
Benchmark harness for stress-testing agentic workflows across diverse collaborative tasks.

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.