Google Gemma 4

Google DeepMind has released Gemma 4, a family of open-weight language models in four sizes (E2B, E4B, 26B MoE, and 31B Dense) under the Apache 2.0 license. The new release brings dramatic benchmark gains over Gemma 3, with AIME 2026 math jumping from 20.8% to 89.2%, LiveCodeBench coding from 29.1% to 80.0%, and GPQA science from 42.4% to 84.3%. The flagship 31B Instruct variant ranks #3 on Arena AI text leaderboard at 1452 Elo, outperforming closed models twenty times its size. Gemma 4 ships with day-one support for Hugging Face, Kaggle, Ollama, and Google Cloud Vertex AI.

Black Forest Labs Drops FLUX.2

Black Forest Labs has launched FLUX.2, an advanced image generation model offering native 4MP output, multi-reference editing, improved prompt adherence, and high-quality typography. It supports any aspect ratio and generates images in under 10 seconds. Key features include superior text rendering, anatomy, and scene composition, along with color matching and efficient VRAM usage. The FLUX.2 lineup includes several models, each tailored for different needs, with varying levels of access and licensing. The minimum system requirements are specified for optimal performance. FLUX.2 is recognized as a leading open-weight image model, suitable for various creative and commercial applications. The model and resources are available on platforms like Hugging Face and GitHub.

OpenAI’s gpt-oss Models: A New Era for Open-Source AI

OpenAI has released two open-weight models, gpt-oss-120b and gpt-oss-20b, under Apache 2.0, marking a shift towards open-source AI. These models are designed for efficiency and performance, with the 120b model featuring 120 billion parameters suitable for a single 80GB GPU and the 20b model running on 16GB of VRAM, accessible for consumer hardware. They excel in reasoning, code generation, and tool integration, competing with leading AI models. Despite not releasing training data due to legal concerns, OpenAI emphasizes the models' performance and encourages community feedback through a Red Teaming Challenge to enhance safety. These models offer developers powerful tools for AI applications without proprietary constraints, fostering innovation in areas like code assistance, customer service, and research tools.