State-of-the-Art / Neural Networks and Deep Learning

Google Launches Gemini 2.5 Flash Image with Text-Based Editing Capabilities

26 August 2025

Google Launches Gemini 2.5 Flash Image with Text-Based Editing Capabilities

Google introduced Gemini 2.5 Flash Image (with internal codename nano-banana) — a model for image generation and editing. The model supports combining multiple images into one, maintains character consistency between…

NVIDIA Nemotron Nano 2: Reasoning and Code Generation Model Outperforms Qwen3-8B on Benchmarks and Supports 128k Context

20 August 2025

NVIDIA Nemotron Nano 2: Reasoning and Code Generation Model Outperforms Qwen3-8B on Benchmarks and Supports 128k Context

A team of NVIDIA researchers presented Nemotron-Nano-9B-v2 — a hybrid Mamba-Transformer language model that generates responses 6 times faster than Qwen3-8B on reasoning tasks while exceeding it in accuracy. The…

Matrix-3D: Open Framework for Generating Fully Explorable 3D Worlds from a Single Image

14 August 2025

Matrix-3D: Open Framework for Generating Fully Explorable 3D Worlds from a Single Image

Researchers from Skywork AI and the Hong Kong University of Science and Technology have introduced Matrix-3D — a framework for creating fully explorable 3D worlds from a single image or…

3D-R1: Open Source Reasoning Model for 3D Scenes Outperforms State-of-the-Art Methods by 10% on 3D Benchmarks

6 August 2025

3D-R1: Open Source Reasoning Model for 3D Scenes Outperforms State-of-the-Art Methods by 10% on 3D Benchmarks

Researchers from Shanghai University of Engineering Science and Peking University presented 3D-R1 — a new foundation model that significantly improves reasoning capabilities in three-dimensional vision-language models (VLM). The model demonstrates an average performance…

Seed Diffusion: New State-of-the-Art in Speed-Quality Balance for Code Generation Models

6 August 2025

Seed Diffusion: New State-of-the-Art in Speed-Quality Balance for Code Generation Models

The research team from ByteDance Seed in collaboration with the AIR Institute of Tsinghua University introduced Seed Diffusion Preview — a language model based on discrete diffusion that demonstrates record-breaking…

Gemini 2.5 Pro Achieved Gold Medal Performance at IMO 2025, Solving 5 of 6 Problems

25 July 2025

Gemini 2.5 Pro Achieved Gold Medal Performance at IMO 2025, Solving 5 of 6 Problems

Large language models perform well on mathematical benchmarks like AIME, however International Mathematical Olympiad (IMO) problems require deep understanding, creativity, and formal reasoning. Chinese researchers used Google Gemini 2.5 Pro…

Show-o2: Open-source 7B multimodal model outperforms 14B models on benchmarks using significantly less training data

11 July 2025

Show-o2: Open-source 7B multimodal model outperforms 14B models on benchmarks using significantly less training data

Researchers from Show Lab at the National University of Singapore and ByteDance introduced Show-o2 — a second-generation multimodal model that demonstrates superior results in image and video understanding and generation…

MiniCPM4: Open Local Model Achieves Qwen3-8B Performance with 7x Inference Acceleration

15 June 2025

MiniCPM4: Open Local Model Achieves Qwen3-8B Performance with 7x Inference Acceleration

The OpenBMB research team presented MiniCPM4 — a highly efficient language model designed specifically for local devices. MiniCPM4-8B achieves comparable performance to Qwen3-8B (81.13 vs 80.55), while requiring 4.5 times…

Strict On-Policy Training with Optimal Baseline: Microsoft Introduces Simplified Algorithm for RLHF

4 June 2025

Strict On-Policy Training with Optimal Baseline: Microsoft Introduces Simplified Algorithm for RLHF

The Microsoft Research team introduced On-Policy RL with Optimal reward baseline (OPO) — a simplified reinforcement learning algorithm for aligning large language models. The new method addresses key problems of…

Visual-ARFT: Multimodal AI Agents Outperform GPT-4o by 18.6% in Complex Visual Tasks

22 May 2025

Visual-ARFT: Multimodal AI Agents Outperform GPT-4o by 18.6% in Complex Visual Tasks

A research team from Shanghai Jiao Tong University and Shanghai Artificial Intelligence Laboratory has introduced Visual Agentic Reinforcement Fine-Tuning (Visual-ARFT) — a new approach to training large multimodal models with…

ZEROSEARCH: A Framework That Cuts LLM Search Training Costs by 88%

9 May 2025

ZEROSEARCH: A Framework That Cuts LLM Search Training Costs by 88%

Alibaba’s NLP research team has officially open-sourced ZEROSEARCH, a complete framework for training LLMs to search without using real search engines. ZEROSEARCH builds on a key insight: LLMs have already…

Phi-4-reasoning: Microsoft’s Breakthrough in AI Thinking

4 May 2025

Phi-4-reasoning: Microsoft’s Breakthrough in AI Thinking

Microsoft recently unveiled Phi-4-reasoning, a 14-billion parameter model that achieves exceptional performance on complex reasoning tasks, outperforming models 5-47 times larger while requiring significantly less computational resources, with developers able…

MedSAM2: Open Source SOTA 3D Medical Image and Video Segmentation Model

13 April 2025

MedSAM2: Open Source SOTA 3D Medical Image and Video Segmentation Model

Medical image segmentation plays a critical role in precision medicine, enabling more accurate diagnosis, treatment planning, and quantitative analysis. While significant progress has been made in developing both specialized and…

Llama Nemotron: NVIDIA Launches Family of Open Reasoning AI Models Overtaking DeepSeek R1

19 March 2025

Llama Nemotron: NVIDIA Launches Family of Open Reasoning AI Models Overtaking DeepSeek R1

NVIDIA has announced the open Llama Nemotron family of models with reasoning capabilities, designed to provide a business-ready foundation for creating advanced AI agents. These models can work independently or…

Chain-of-Experts: Novel Approach Improving MoE Efficiency with up to 42% Memory Reduction

11 March 2025

Chain-of-Experts: Novel Approach Improving MoE Efficiency with up to 42% Memory Reduction

Chain-of-Experts (CoE) – a novel approach fundamentally changing how sparse language models process information, delivering better performance with significantly less memory. This breakthrough addresses key limitations in current Mixture-of-Experts (MoE)…

R1-Onevision: Open Source 7B-Parameter Model Outperforming GPT-4o in Maths and Reasoning Tasks

27 February 2025

R1-Onevision: Open Source 7B-Parameter Model Outperforming GPT-4o in Maths and Reasoning Tasks

Researchers from Zhejiang University have released R1-Onevision, a 7B parameters multimodal reasoning model that processes and analyzes visual inputs with unprecedented logical precision, capable of understanding complex mathematical, scientific, and…

Step-Video-T2V: Text-to-Video Open-Source Model Achieves 16x Video Compression Breakthrough

20 February 2025

Step-Video-T2V: Text-to-Video Open-Source Model Achieves 16x Video Compression Breakthrough

Researchers from Stepfun AI have developed Step-Video-T2V, a 30-billion-parameter text-to-video model that generates videos up to 204 frames in length, 544×992 resolution, capable of understanding both Chinese and English prompts.…

EPFL Study: Language Models Don’t Translate Into English – They Operate Through Concepts

30 January 2025

EPFL Study: Language Models Don’t Translate Into English – They Operate Through Concepts

New research from EPFL sheds light on the internal mechanisms of multilingual data processing in LLMs, which is critical for understanding how modern language models work and how to optimize…

ByteDance Unveil TA-TiTok Tokenizer Achieving SOTA in Text-to-Image Generation Using Only Public Data

19 January 2025

ByteDance Unveil TA-TiTok Tokenizer Achieving SOTA in Text-to-Image Generation Using Only Public Data

Researchers from ByteDance and POSTECH introduced TA-TiTok (Text-Aware Transformer-based 1-Dimensional Tokenizer), a novel approach to making text-to-image AI models more accessible and efficient. Their work demonstrates through MaskGen models how…

MiniMax-01: 4M Context Length Benchmark Leader Powered by Lightning Attention

15 January 2025

MiniMax-01: 4M Context Length Benchmark Leader Powered by Lightning Attention

MiniMax has open-sourced its latest MiniMax-01 series, introducing two models that push the boundaries of context length and attention mechanisms: MiniMax-Text-01 for language processing and MiniMax-VL-01 for visual-language tasks. The…

ArtAug: Open Source Framework for Image Generation Models Enhancement

18 December 2024

ArtAug: Open Source Framework for Image Generation Models Enhancement

East China Normal University and Alibaba Group researchers have introduced ArtAug, a framework that enhances text-to-image generation through synthesis-understanding interaction. This approach significantly improves image quality without requiring extensive manual…