A new generation of voice AI that listens and speaks simultaneously, rolling out to ChatGPT on iOS, Android, and web with API support coming soon.
Trained on tens of thousands of NVIDIA GB300 GPUs with large-scale RL, served at 80 TPS, and priced at $2/M input and $6/M output tokens. Available in Grok Build, Cursor, and the SpaceXAI API.
The AI chip maker for enterprise inference infrastructure drew a $1 billion round led by General Atlantic with Intel participating.
Google introduces open-weight multimodal LLMs (2.3B-31B, dense + MoE) with thinking mode, encoder-free 12B architecture, and KV-cache optimizations, establishing SOTA on STEM, multimodal, and long-context benchmarks.
NVIDIA's tri-mode LM at 8B scale decodes 6x more tokens per forward than Qwen3-8B with comparable accuracy, delivering 4x higher throughput on SPEED-Bench.
Formulates detection, segmentation, depth, keypoints and pose as multimodal generation within a single UMM without task-specific heads, matching specialized systems across diverse vision benchmarks.
Jointly predicts synchronized RGB, depth and optical flow from a single RGB-D image and language instruction. Features a tri-branch Diffusion Transformer enabling >9 Hz closed-loop bimanual control.
A unified video diffusion framework enabling flexible generation with both forward and bidirectional conditioning in a single architecture.
Analysis from Theo — t3.gg benchmarks Grok 4.5 against leading models, showing competitive performance at a fraction of the cost with significantly better token efficiency.
Bloomberg's AI editor reports on the global rollout of OpenAI's most advanced model following regulatory clearance.
Bloomberg sources reveal BofA is now extending a $520 million credit line to OpenAI after initially passing on the opportunity, signaling growing financial confidence.
French AI startup ZML released a free LLM inference server supporting multi-chip deployment (Nvidia, AMD, Google TPU, Apple Metal, Intel Arc), promising maximum performance without vendor lock-in.
An AI coworker with dedicated work surfaces for email, meetings, notes, browser, and coding. Reached #2 on Show HN with 168 points.
Generates verifiable audit trails proving what an agent did with data, aiming to solve the compliance and trust gap in the post-agentic enterprise.
AI-powered design, writing, analysis, and organization tools on the Framer canvas, plus branching for safer team collaboration.
Targets high-accuracy transcription use cases across multiple languages and domains with improved precision.
The Sapienza University of Rome spin-off secured €1.85M pre-seed to develop an enterprise AI governance and control platform. Founded by AI/NLP researchers from Sapienza, the startup aims to help regulated enterprises safely adopt LLMs and generative AI — exactly the kind of Italian deep tech that deserves the spotlight.
With 79,000 companies in the supply chain and a market projected beyond €117 billion, Italian logistics continues to grow despite geopolitical tensions. AI, data, automation, and new commercial skills are key to turning uncertainty into competitive advantage.
Source:
