16 / Apr 14 – 20, 2025 · 11 RELEASES

AI News Week 16, 2025

Week 16 2025 AI releases.

Language Models (3)

Qwen3
Alibaba Qwen Team · Reasoning
Qwen3 introduces a hybrid thinking/non-thinking mode architecture with models ranging from 0.6B to 235B parameters (MoE), supporting 119 languages and achieving competitive performance on coding, math, and general benchmarks against o1, DeepSeek-R1, and Gemini-2.5-Pro.
Phi Small Language Models
Microsoft · Small Language Models
Microsoft's Phi family of small language models has achieved significant performance improvements over the past year, demonstrating that smaller models can match larger counterparts on key benchmarks through specialized training and optimization techniques.
Absolute-Zero-Reasoner
LeapLabTHU · Reasoning
A reasoning framework that appears to focus on inference-time computation optimization for language models, enabling more efficient reasoning capabilities without additional training.

ICEdit
Zhejiang University & Harvard University · Image Editing
Instruction-based image editing framework using Diffusion Transformers with in-context prompting, requiring only 1% trainable parameters (200M) and 0.1% training data (50k) compared to existing methods. Achieves competitive results with commercial models like Gemini and GPT-4o while being faster (~9 seconds per image) and more efficient on consumer hardware (4GB VRAM).

FantasyTalking
Alibaba AMAP & Beijing University of Posts and Telecommunications · Text-to-Video
A video diffusion transformer framework that generates realistic talking portrait videos from static images using dual-stage audio-visual alignment for coherent lip-sync, facial expressions, and body motion with controllable motion intensity. Supports diverse character styles and body ranges while preserving identity consistency.
3DV-TON
Alibaba DAMO Academy, Hupan Lab, Zhejiang University · Text-to-Video
A diffusion-based framework for generating high-fidelity video try-on by using animatable textured 3D meshes as explicit frame-level guidance to maintain temporal consistency and accurate clothing details across complex poses. Introduces HR-VVT, a high-resolution benchmark dataset with 130 diverse videos.

Suno v4.5
Suno · Music Generation
Music generation model update with improvements to audio quality, composition consistency, and user control over generated tracks.

DesktopCommanderMCP
wonderwhy-er · Agent Framework
An MCP (Model Context Protocol) server that enables AI agents to control desktop applications and execute system commands through a standardized interface. Allows LLMs to interact with GUI applications and perform automation tasks on local machines.

Synthetic Data Kit
Meta · Data Integration
Meta's toolkit for generating synthetic data to train and improve AI models, providing utilities for data creation, augmentation, and quality assessment without requiring large labeled datasets.
EdgeTAM
Facebook Research · Edge Deployment
Edge-optimized Tracking and Mapping system that enables efficient visual tracking and spatial mapping on resource-constrained devices. Designed for real-time performance on edge hardware while maintaining accuracy comparable to server-based solutions.
Gemini Cookbook
Google · Prompt Engineering
Collection of practical examples and tutorials for building applications with Google's Gemini API, covering multimodal inputs, function calling, and various integration patterns.

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