Open source / Neural Networks and Deep Learning

MinerU-Diffusion: A New Approach to OCR via Diffusion Decoding Speeds Up PDF Parsing 3× Without Accuracy Loss

27 March 2026

MinerU-Diffusion: A New Approach to OCR via Diffusion Decoding Speeds Up PDF Parsing 3× Without Accuracy Loss

A team from Shanghai Artificial Intelligence Laboratory and Peking University published MinerU-Diffusion — a document OCR framework that abandons classical autoregressive generation in favor of diffusion-based decoding. The project is…

daVinci-MagiHuman: Open 15B Model Generates a 5-Second Lip Sync Video in 2 Seconds on a Single H100

24 March 2026

daVinci-MagiHuman: Open 15B Model Generates a 5-Second Lip Sync Video in 2 Seconds on a Single H100

SII-GAIR and Sand.ai have published daVinci-MagiHuman — an open-source multimodal 15B model based on a single-stream transformer that simultaneously generates video with precise lip sync and synchronized audio, producing a…

OpenClaw: The Lobster That Took Over the World — How One Developer Built the Most Popular Open-Source AI Agent in History

18 March 2026

OpenClaw: The Lobster That Took Over the World — How One Developer Built the Most Popular Open-Source AI Agent in History

OpenClaw is a free and open-source AI agent created by Austrian developer Peter Steinberger in November 2025. An AI agent is a software wrapper around a language model that does…

OpenClaw-RL: A Framework That Updates an Agent’s Weights on the Fly, Learning from User and Environment Feedback

17 March 2026

OpenClaw-RL: A Framework That Updates an Agent’s Weights on the Fly, Learning from User and Environment Feedback

Researchers from Princeton University have introduced OpenClaw-RL, a framework that allows an AI agent to improve in real time — without a separate data collection stage and without manual annotation.…

Helios: 14B Model Generates Videos Longer Than 60 Seconds at 19.5 FPS on a Single H100

11 March 2026

Helios: 14B Model Generates Videos Longer Than 60 Seconds at 19.5 FPS on a Single H100

A team of researchers from Peking University and ByteDance published Helios — an autoregressive diffusion transformer with 14 billion parameters that generates video at 19.5 frames per second on a…

VBVR: 2 Million Videos for Reasoning Training — an Open Dataset That Changes the Rules

26 February 2026

VBVR: 2 Million Videos for Reasoning Training — an Open Dataset That Changes the Rules

A team of more than 50 researchers from around the world — from Berkeley, Stanford, CMU, Oxford and other universities — has published Very Big Video Reasoning (VBVR), a massive…

GLM-5: Top-1 Open-Weight Model for Code and Text Generation, Competing with Claude and GPT on Agentic Tasks

19 February 2026

19 February 2026

GLM-5: Top-1 Open-Weight Model for Code and Text Generation, Competing with Claude and GPT on Agentic Tasks

19 February 2026

Zhipu AI and Tsinghua University have published a GLM-5 technical report — currently the top-performing open-weight language model by benchmarks: first place among open-weight models on Artificial Analysis and top-1…

Baichuan-M3: An Open Medical Model That Conducts Consultations Like a Real Doctor and Outperforms GPT-5.2 on Benchmarks

10 February 2026

Baichuan-M3: An Open Medical Model That Conducts Consultations Like a Real Doctor and Outperforms GPT-5.2 on Benchmarks

A research team from the Chinese company Baichuan has introduced Baichuan-M3 — an open medical language model that, instead of the traditional question-and-answer mode, conducts a full clinical dialogue, actively…

Claude Sonnet 4.5 Leads on Comprehensive Backend Benchmark, Outperforming in Both Code and Environment Configuration

22 January 2026

Claude Sonnet 4.5 Leads on Comprehensive Backend Benchmark, Outperforming in Both Code and Environment Configuration

A team of researchers from Fudan University and Shanghai Qĳi Zhifeng Co. introduced ABC-Bench — the first benchmark that tests the ability of AI agents to solve full-fledged backend development…

Multiplex Thinking: Sampling 3 Tokens Instead of 1 Increases Olympiad Problem-Solving Accuracy from 40% to 55%

22 January 2026

Multiplex Thinking: Sampling 3 Tokens Instead of 1 Increases Olympiad Problem-Solving Accuracy from 40% to 55%

Researchers from the University of Pennsylvania and Microsoft Research introduced Multiplex Thinking — a new reasoning method for large language models. The idea is to generate not one token at…

Yume1.5: An Open Model for Creating Interactive Virtual Worlds with Keyboard Control

5 January 2026

Yume1.5: An Open Model for Creating Interactive Virtual Worlds with Keyboard Control

Researchers from Shanghai AI Laboratory and Fudan University published Yume1.5 — a model for generating interactive virtual worlds that can be controlled directly from the keyboard. Unlike regular video generation,…

Wan-Move: Open-Source Alternative to Kling 1.5 Pro for Motion-Controllable Video Generation

13 December 2025

Wan-Move: Open-Source Alternative to Kling 1.5 Pro for Motion-Controllable Video Generation

A team of researchers from Tongyi Lab (Alibaba Group), Tsinghua University, and the University of Hong Kong presented Wan-Move — a new approach to precise motion control in generative video…

P1: First Open-Source Model to Win Gold at the International Physics Olympiad

30 November 2025

P1: First Open-Source Model to Win Gold at the International Physics Olympiad

P1-235B-A22B from Shanghai AI Laboratory became the first open-source model to win a gold medal at the latest International Physics Olympiad IPhO 2025, scoring 21.2 out of 30 points and…

MiroThinker v1.0: Open-Source AI Research Agent Learns to Make Up to 600 Tool Calls Per Task

20 November 2025

MiroThinker v1.0: Open-Source AI Research Agent Learns to Make Up to 600 Tool Calls Per Task

The MiroMind team introduced MiroThinker v1.0 — an AI research agent capable of performing up to 600 tool calls per task with a 256K token context window. On four key…

DeepEyesV2: Multimodal Model Learns to Use Tools to Solve Complex TasksRetry

12 November 2025

DeepEyesV2: Multimodal Model Learns to Use Tools to Solve Complex TasksRetry

Researchers from Xiaohongshu introduced DeepEyesV2 — an agentic multimodal model based on Qwen2.5-VL-7B that can not only understand text and images but also actively use external tools: execute Python code…

Remote Labor Index: Top AI Agents Successfully Complete 2.5% of Freelance Projects

4 November 2025

4 November 2025

Remote Labor Index: Top AI Agents Successfully Complete 2.5% of Freelance Projects

4 November 2025

A team of researchers from the Center for AI Safety and Scale AI published the Remote Labor Index (RLI) — the first benchmark that tests whether AI agents can perform…

Ditto: Open Framework for Text-Instruction-Based Video Style and Object Editing with 99% Frame Consistency

24 October 2025

Ditto: Open Framework for Text-Instruction-Based Video Style and Object Editing with 99% Frame Consistency

Researchers from HKUST, Ant Group, Zhejiang University, and Northeastern University introduced Ditto — a comprehensive open framework addressing the training data scarcity problem in text-instruction-based video editing. The developers created…

QeRL: Training 32B Models on Single H100 vs Three GPUs, Beating LoRA in Accuracy

16 October 2025

QeRL rainforcement learning quantization training speedup

QeRL: Training 32B Models on Single H100 vs Three GPUs, Beating LoRA in Accuracy

QeRL is a framework for training language models using reinforcement learning that simultaneously reduces GPU requirements and surpasses traditional LoRA and QLoRA methods in accuracy. On the Qwen2.5-7B-Instruct model, QeRL…

Kimi-K2 and Qwen3-235B-Ins – Best AI Models for Stock Trading, Chinese Researchers Found

10 October 2025

10 October 2025

Kimi-K2 and Qwen3-235B-Ins – Best AI Models for Stock Trading, Chinese Researchers Found

10 October 2025

Researchers from China conducted a large-scale comparison of AI capabilities for stock trading using real market data. AI agents managed a portfolio of 20 Dow Jones Index stocks over 4…