Runway has launched a model router that selects the optimal image, video, or audio generation model for each API request. Developers set their priority — quality, latency, or price — and the system handles the rest.
A new open-source tokenizer leverages GPU parallelism to process text roughly 1000x faster than CPU-based alternatives like HuggingFace Tokenizers — a meaningful shift for high-throughput LLM pipelines.
A fabricated image showing Senator Mitch McConnell in apparent medical distress spread online before Google's deepfake detection system helped confirm it was AI-generated. Here's why that matters for builders working on media verification.
Meta's experimental app Pocket lets users type a text prompt and get a shareable, interactive mini game in return — a direct consumer bet on AI-generated interactive experiences.
Google has released a streamlined version of its Imagen 2 model designed to cut generation time and API costs — a practical upgrade for developers and creators building AI-powered visual workflows.
The open-source agentic AI program OpenClaw has launched mobile apps for both Android and iOS, bringing autonomous task execution to your phone for free.
Cursor's new mobile app lets developers monitor and direct AI coding agents away from their desk, extending agent oversight to phones.
As more discovery shifts from search engines to LLMs, a new site called In the Weights lets you check how strongly frontier and smaller models recognize your name or work.
Anthropic's Claude SDK now officially supports Apple Foundation Models, letting developers run Claude-compatible workflows against on-device Apple intelligence without routing every call to the cloud.
Anthropic's latest model generates functional, genuinely entertaining browser games from a plain-English description. Prototyping interactive experiences no longer requires a game engine or a dev team.
Apple has published official documentation for Core AI, a new first-party framework that lets developers run machine learning models directly on Apple hardware — no cloud required.
Cognition AI has released FrontierCode, a benchmark designed to measure how well AI coding agents handle complex, real-world software engineering tasks beyond simple autocomplete or isolated puzzles.
