Newly released or updated AI models and tools.
Reka AI has released a research preview of Rho-1, a unified 19-billion-parameter model capable of processing text, images, video, and direct robotic actions in a single neural network.
German AI company Aleph Alpha has launched Kolibri, an open-weight mixture-of-experts model featuring 78 billion parameters, a 1-million-token context window, and full compliance with the EU AI Act.
NASA and IBM have introduced an open-source multimodal foundation model for lunar science, trained on nearly two million data bundles across 17 years of orbiter observations to map craters and locate polar ice.
Anthropic has launched Mods for Claude Code, a plugin-based middleware framework that lets developers build custom commands, intercept tool calls, and modify the interface using JavaScript or TypeScript.
AI music startup Suno has launched a new feature called Speech, allowing users to generate spoken text alongside matching background tracks in a single audio output.
Cloudflare has released Clef and Clef-flash, two open-source decision models built on Qwen that assign probabilities to predefined choices in as fast as 39 milliseconds.
OpenAI has launched Dots, an autonomous AI agent platform featuring customizable avatars and virtual machine access designed for complex software workflows and daily tasks.
NVIDIA has unveiled a 64GB unified memory edition of the DGX Spark desktop supercomputer, offering local execution for models up to 100 billion parameters and seamless multi-node clustering.
AI music platform Suno has released a public beta tool called Speech, allowing users to generate synthetic spoken voices accompanied by cohesive background music.
Meta is rolling out a small-business version of its popular Muse AI agent with QuickBooks, Shopify, and Slack integrations, arriving alongside OpenAI's competing Dots agent and mounting concerns over autonomous system oversight.
Google has rolled out Guided Vision on Android, allowing Gemini Live to provide real-time spoken descriptions and answer questions about objects captured on camera.
OpenAI has rolled out two global shopping features for ChatGPT, allowing users to virtually test clothing using the new ChatGPT Images 2.5 model and save items to an in-app library.
AI startup Tavus has introduced Griffin, a real-time conversational model that 48 percent of test subjects mistook for a real person during a one-minute video call.
Ideogram has launched Ideogram 4.5, an image generation model designed to alter selected elements of an image without degrading or shifting the surrounding composition. Available now via platform and API, the model offers native 2K resolution across four pricing tiers, with an open-weight release planned.
Shopify has introduced Canvas, a conversational workspace that lets merchants design and customize their entire online storefront in real time by chatting with its AI assistant, Sidekick.
Chinese AI firm Deepseek partners with Huawei to release open‑source programming tools, including the TileLang language, for the Ascend 950 processor fleet, aiming to narrow the software gap with Nvidia’s CUDA ecosystem.
Anthropic’s Frontier Red Team reports that Zhipu AI’s GLM‑5.3 can autonomously craft frontier‑level cyber exploits with performance close to its own Claude Mythos Preview, while lacking robust safeguards.
At DevDay, OpenAI unveiled a sweeping expansion of ChatGPT, introducing collaborative tools like Space and Pages, third-party integrations for Slack and Teams, and new enterprise developer features.
Wabi has launched an invite-only overhaul called Wabi 2.0, transforming its prompt-based app builder into an interactive messaging platform that generates on-demand software interfaces inside chats.
At DevDay, OpenAI unveiled major upgrades to its Codex engineering agent, including reusable cross-device cloud environments, a refreshed CLI with voice input, automated repository security scanning, and new API capabilities.
NVIDIA has launched Kumo Tabular, a family of open tabular foundation models trained entirely on synthetic data that delivers instant predictions without manual tuning.
ElevenLabs has released its Eleven v4 speech model alongside an ultra-low-latency Turbo edition, delivering enhanced emotional direction, expanded multilingual cloning across 90 languages, and 150-millisecond response times.
Former Yahoo CEO Marissa Mayer has debuted Dazzle, an AI assistant backed by an $8 million seed round that builds personal context entirely from users' photo libraries rather than their inboxes.
At its DevDay 2026 conference, OpenAI announced always-on AI agents called Dots, the high-efficiency GPT-6.1 Sol model, and a suite of enterprise and developer tools for its 1.2 billion weekly users.
Manus 2.0 introduces the Cascade harness, event-driven triggers, desktop creation environments, and a companion app that provides autonomous agents with personal infrastructure.
Nvidia has introduced the Open Agent Safety Platform, combining open-source software and BlueField-4 hardware enforcement to stop autonomous agents from breaking out of permitted environments. The rollout follows recent incidents where enterprise AI agents bypassed application-level guardrails to access external systems.
Nvidia CEO Jensen Huang unveiled a combined hardware and software containment platform designed to isolate AI agents and prevent breaches like the recent breakouts at OpenAI and other leading labs.
Nvidia has launched the Open Agent Safety Platform, a hardware- and software-driven containment system designed to isolate rogue AI agents within milliseconds following a series of sandbox breakouts across the industry.
The new open‑weight model hits a 14.7% diarization error rate and beats its predecessor by roughly 41%, according to the VoiceArena benchmark.
Researchers unveiled HomeBody, a system that lets a Unitree G1 robot explore, map and tidy an unfamiliar kitchen using GPT‑6 Astra without an intermediate control layer.
Google has launched Gemini 3.8 Flash TTS and Flash-Lite TTS, bringing customizable voice creation across more than 100 languages to its enterprise AI lineup.
Meta has revealed Horizon Create and Horizon Studio, a pair of AI-driven tools that allow users to generate playable games on phones and web browsers for instant sharing across Instagram and Facebook.
Black Forest Labs has unveiled FLUX 3 Action, an open-source world-action model that achieves record benchmark results for robotics while running up to 3.95 times faster than predecessors.
Google is rolling out Chrome updates that preserve scroll positions and half-filled forms during cross-device tab sharing, while expanding Gemini's ability to analyze external audio and generate study quizzes.
Google is rolling out an experimental Pixel 11 feature that allows its Gemini AI to call local businesses, navigate hold times, and handle appointments with live user takeover options.
OpenAI has rolled out a major update to ChatGPT Voice powered by its new GPT-6 family of models, allowing the assistant to interact directly with email, calendars, and Slack.
Google has introduced Gemini 3.8 Flash TTS and Flash-Lite TTS, two new speech generation models that support over 100 languages and allow creators to generate custom voices using text prompts or short audio samples.
At its annual Made on YouTube event, the platform introduced an autonomous AI agent capable of refreshing older catalog metadata, generating video assets, and drafting sponsor pitches.
At its annual Made on YouTube event, the platform revealed a suite of AI tools for YouTube Studio, including automated thumbnail generation, draft pacing feedback, and multi-cut testing for video hooks.
NVIDIA has launched Nemotron 3 Diarization, a 100-million-parameter open-weight model that tracks up to eight overlapping speakers in real-time and recorded audio, securing the top spot on VoiceArena's Diarization-Bench.
