Google Labs introduced Playground, an experimental platform where people build their own games by chatting with AI, then play and share them without writing code. You can start from genres like Trivia, Tower Defense and Racing or from scratch. It is open to users 18 and over in the US for now.
llama.cpp
llama.cpp, the open-source program for running AI models on your own computer, has released version 0.6.0. Its server gains /v1/systemone, an API for decision models, and Clef is now supported with text and images. The release also extends speed-up methods such as MTP and DFlash, adds GLM-5.3-Flash
Reflection AI
Reflection AI introduced Beam, its first open-weight model, with 501 billion total and 23 billion active parameters, trained for coding, reasoning and agentic work. According to the company, the weights will be released this month under Apache 2.0; for now, early access runs through a waitlist.
Bilibili
Bilibili's Index team has released Index-Translate, a family of translation models built on Qwen3.5, under the Apache 2.0 license. The text models cover 150 languages, including Turkish and Arabic, and the family also includes models for speech translation, syllable-controlled dubbing and long docum
GitHub
GitHub released two Copilot features in preview. With computer use, Copilot can act in desktop apps on macOS and Windows after asking for approval. Dynamic workflows combine automated steps and agent work in a single program.
Microsoft
Microsoft AI has put MAI-Transcribe-2-Streaming, its first model that transcribes speech while the speaker talks, into public preview. It supports 60 languages, including Turkish, with automatic language detection. The company also introduced two new voice models for voice agents, MAI-Voice-2.1 and
Black Forest Labs
Black Forest Labs has launched FLUX 3 Image, the image model of its FLUX 3 family, through its API. It lets users place elements with bounding boxes, keeps untouched pixels exactly the same during edits, combines up to ten references and outputs up to 4K. The company says an open-weights version is
AWS
Strands Labs, where AWS gathers experimental agent projects, released Strands Decider 2B, a small open decision model. With about 1.9 billion parameters, it picks between options and attaches a confidence score to every decision, and it runs on NVIDIA GPUs, Apple silicon Macs and CPU-only machines.
Cloudflare
Cloudflare introduced Clef and Clef-flash, the first models trained in-house by its Workers AI team. These decision models pick from predefined options and return a probability for each one. They run on Workers AI and are available on Hugging Face, and Cloudflare is also launching a reinforcement le
Ai2
Ai2 has released AstaBrief 8B under the Apache 2.0 license, an 8-billion-parameter model that turns a research question and retrieved paper excerpts into a cited report. Built on Qwen3-8B, it powers the new Fast mode in the Asta platform and can also be run locally.
Claude Code
Anthropic has added mods to Claude Code. Written in JavaScript or TypeScript, mods can change tool calls, prompts and the interface, and some built-in features such as /diff are built as mods. Mods are not sandboxed, so Anthropic advises installing them only from trusted sources.
Aleph Alpha
Aleph Alpha has released Kolibri, an open-weight model for English and German, under the Apache 2.0 license. The model has 78.1 billion parameters in total with 3.46 billion active per token, reads up to 1 million tokens of context and is trained to say it does not know when the context has no answe