Google expands its Gemini API with powerful new features for RAG systems, including multimodal search for images and text, custom metadata filtering for faster and more accurate queries, and page-level citations that link AI responses directly to their source documents.
Google releases open-source Multi-Token Prediction drafters for Gemma 4 models, tripling AI inference speed through speculative decoding with zero quality loss, now freely available on Hugging Face and Kaggle under an Apache 2.0 license.
Rapid-MLX launches as the fastest local AI engine for Apple Silicon, running large language models up to 4.2x faster than Ollama with a Time to First Token as low as 0.08 seconds, offering full OpenAI API compatibility and support for tools like Cursor, Claude Code, and models ranging from compact …
IBM, Cleveland Clinic, and Riken achieve a quantum computing breakthrough by simulating a 12,635-atom protein complex — a massive leap from just 10 atoms in October 2024 — unlocking powerful new possibilities for drug discovery and disease treatment.
OpenAI launches GPT-5.5 Instant with 52% fewer hallucinations, replacing GPT-5.3 Instant as ChatGPT's default model while rolling out new memory controls across all ChatGPT models for greater user transparency and personalization.
OpenAI and major tech giants including Microsoft, Nvidia, AMD, and Intel unveil MRC, a groundbreaking open networking protocol using 'packet spraying' to dramatically boost AI GPU cluster speed, reliability, and energy efficiency — already live in major supercomputing sites — as the industry races to combat a looming compute shortage.
Voice-Pro goes fully open-source, giving everyone free access to powerful AI voice cloning, speech recognition, multilingual text-to-speech, and vocal isolation tools as its developers shift focus to a new global communication platform.
Vercel open sources deepsec, an AI-powered security tool that scans codebases for vulnerabilities directly on users' own infrastructure, leveraging existing Claude or Codex subscriptions with a multi-step workflow that reduces false positives to just 10-20% and supports 1,000+ concurrent sandboxes for large repositories.
Facebook Research unveils Tuna-2, a groundbreaking multimodal AI model that ditches traditional vision encoders in favor of direct pixel patch processing, outperforming predecessors on diverse benchmarks while supporting both image understanding and generation tasks in 7B and 2B parameter sizes.
DeepSeek unleashes two groundbreaking AI models — DeepSeek-V4-Pro and V4-Flash — featuring over 1 trillion parameters and record-breaking low prices, with V4-Flash undercutting OpenAI's cheapest model at just $0.14 per million tokens, powered by dramatic efficiency gains that slash compute costs to a fraction of its predecessor.