ByteDance launched PaSa, an intelligent academic paper search agent. It aims to solve the problems of complex query handling in academic research and helps researchers save time in literature retrieval.
MoreLatest AI Events
ByteDance Unveils PaSa: An AI-Powered Academic Paper Search Solution


OpenAI launches Operator, an AI agent for automated web tasks
OpenAI has introduced Operator, a new AI agent capable of autonomously performing tasks in a web browser. Powered by the Computer-Using Agent (CUA) model, Operator can handle activities like booking tickets, managing grocery orders, and more. Initially available to ChatGPT Pro subscribers, it aims to enhance productivity by allowing users to delegate complex online tasks. OpenAI plans to expand access to other subscription tiers and is collaborating with various companies to ensure compliance with service agreements
MoreIntroducing Spell, a model to generate 3D worlds

Spell can generate entire 3D scenes or “Worlds” from an image in just a few minutes. The worlds are consistent with the initial image input and are represented as a volume that can be rendered using Gaussian Splatting (or other methods, like NeRFs).
More
Bytedance lanches Seed Edge for frontier AGI research
In late January, ByteDance officially launched a research project codenamed "Seed Edge," with the core objective of conducting long-term, foundational AGI (Artificial General Intelligence) frontier research that goes beyond pre-training and large model iterations. Seed Edge has already outlined five major research directions.
MoreByteDance Launches Doubao-1.5-Pro, Surpassing GPT-4o in Key Benchmarks

ByteDance released Doubao-1.5-Pro, an advanced AI model utilizing a sparse MoE architecture, achieving performance comparable to dense models with 7x fewer activation parameters. It outperformed GPT-4o, Claude 3.5 Sonnet, and others in coding, reasoning, and Chinese language benchmarks. The model also features enhanced visual and voice capabilities, offering cost-effective solutions for developers.
More
Updated Gemini 2.0 Flash Thinking Experimental model now available
On January 22, 2025, Google unveiled the upgraded Gemini 2.0 Flash Thinking model, designed for complex reasoning tasks. This updated version features a 1-million-token context window and native code execution support, significantly enhancing its analytical capabilities. It aims to address challenges in various fields such as education and research, allowing users to process extensive datasets while maintaining logical consistency. The model's improvements promise to transform industries reliant on advanced AI reasoning.
MoreRelease of Kimi k1.5 Multimodal Thinking Model
The Kimi k1.5 multimodal thinking model achieves state-of-the-art performance in both long-CoT and short-CoT reasoning tasks, matching or surpassing OpenAI's o1 model and outperforming GPT-4o and Claude 3.5 by up to 550% in various benchmarks. Key features include enhanced mathematical, coding, and multimodal reasoning abilities, as well as innovative reinforcement learning techniques that allow for autonomous expansion of training data through a reward mechanism.
MoreOpenAI, Oracle, and SoftBank launch Stargate, a $500 billion AI initiative.

On January 21, 2025, OpenAI, Oracle, and SoftBank announced the Stargate Project, a groundbreaking initiative to invest up to $500 billion in AI infrastructure across the U.S. The project aims to establish data centers and enhance electricity generation in Texas, starting with an initial investment of $100 billion. This venture is expected to create over 100,000 jobs and reinforce U.S. leadership in AI technology amidst global competition, particularly with China.
More
DeepSeek-R1, an OpenAI o1 contender but is 95% cheaper

DeepSeek-R1, an open-source reasoning-focused AI model, was launched on January 20, 2025. It claims to outperform OpenAI's o1 in various benchmarks, particularly in mathematics and coding tasks. With a unique mixture-of-experts architecture and self-verification capabilities, DeepSeek-R1 aims to democratize access to advanced AI technologies at a fraction of the cost of its competitors, fostering innovation and accessibility in AI applications
More
MiniMax unveils MiniMax-01 series, featuring advanced AI models.

Chinese AI startup MiniMax has launched the MiniMax-01 series, which includes MiniMax-Text-01, a text-only model with a 4-million-token context window, and MiniMax-VL-01, a multimodal model. These models aim to compete with industry leaders like OpenAI and Google, boasting significant advancements in processing capabilities and affordability. MiniMax has raised $850 million in funding, positioning itself as a formidable player in the AI landscape despite recent challenges in the global market.
More