Groundbreaking open-source vision-language models GLM-4.5V and GLM-4.1V-Thinking revolutionize multimodal reasoning, achieving state-of-the-art performance on 42 benchmarks and outperforming larger models through Thinking Mode and Chain-of-Thought reasoning.
OpenAI unveils gpt-oss-120b and gpt-oss-20b, open-source large language models that incorporate architectural advances, enable adjustable reasoning, and can run locally while showing competitive performance on benchmarks.
AI model Perch empowers conservationists to safeguard endangered species through improved bioacoustic analysis, recognizing specific sounds, adapting to new environments like coral reefs, and allowing open-source monitoring of biodiversity.
Alibaba unveils Qwen-Image, a powerful open source AI image generator that excels at accurately rendering text within visuals, supporting English and Chinese scripts under the permissive Apache 2.0 license.
OpenAI unleashes two new open source language models gpt-oss-120b and gpt-oss-20b, matching or exceeding their proprietary models' performance while ensuring privacy by running locally, available for free use under the permissive Apache 2.0 license.
Google's AI bug hunter Big Sleep, developed by DeepMind and Project Zero, uncovers 20 security vulnerabilities in widely used open source software like FFmpeg and ImageMagick, marking a milestone in automated vulnerability discovery through artificial intelligence.
The U.S. prioritizes open-source AI as a national imperative to outpace China in the global AI race, recognizing the need to return to open science roots and empower the next generation of AI builders.
Krea unveils open weights for its FLUX.1 image model, promising superior aesthetics and image quality by curating data and calibrating outputs to address the 'AI look', with plans to expand capabilities and offer personalized aesthetics.
Wan2.2, a new video diffusion model, unleashes cinematic video generation by combining Mixture-of-Experts architecture, meticulously curated aesthetic data, and an open-sourced 5B model with high-compression VAE for efficient 720P video synthesis.
A Brazilian prompt engineer achieves a mere 7.5% score on a challenging new AI coding test, underscoring the complexity of real-world coding tasks for artificial intelligence models and prompting the organizer to offer a $1 million prize for an open-source model exceeding 90%.