Windsurf launches SWE-1 AI model for real-time on-device use

Windsurf's Wave 9 SWE-1 is a transformer-based AI model optimized for low-latency, on-device use without cloud dependency, offering real-time multimodal capabilities tailored for privacy-focused, mobile-first applications.

🗞 #windsurf

OpenAI rolled out Codex for automated coding tasks in ChatGPT Pro

OpenAI's Codex is a cloud-based software agent for automating coding tasks like bug fixes and feature implementation. Available May 16, 2025, for select ChatGPT tiers, it operates securely within user codebases with verifiable outputs.

🗞 #chatgpt

Google prepares to launch Flow, a new video editing tool, at I/O 2025

Google is set to introduce Flow, a video generation tool likely debuting at Google I/O, built on updated Veo and Imagen models. Flow appears to revive the Storyboard concept, guiding users from prompt to structured video output.

🗞 #aitestkitchen

Google readies upgrade to Stream Realtime feature in AI Studio

Google quietly updated AI Studio, hinting at upcoming support for real-time multimodal processing via a potential Flash 2.5 model. Backend changes and developer polls suggest an evolving agent system for coding and deployment via Cloud Run.

🗞 #aistudio

First look into upcoming AI-generated Video Overviews from Google

Google's Illuminate project hints at a major shift toward AI-generated multimedia, including audio summaries of texts and short, fully generated video overviews from prompts, powered by a unified model likely tied to Gemini or Veo technologies.

🗞 #aitestkitchen

Google launches NotebookLM mobile app with audio-first features on mobile

Google is launching the NotebookLM mobile app, bringing its web-based features to iOS and Android with a focus on audio content. The app supports podcast-style summaries, interactive playback, and flexible knowledge consumption on the go.

🗞 #notebooklm

Google launches coding agent Jules in beta with free daily tasks

Google's coding agent Jules is now in global beta, letting developers assign tasks directly from GitHub issues. Powered by Gemini 2.5 Pro, it automates pull request workflows and includes a free tier of five tasks per day.

🗞 #ai

Google expected to add credit system to Flow AI video editor

Google's Flow is a modular video creation tool in early testing, offering text, frame, and asset-based inputs. Powered by Veo models and Gemini, it uses a credit system, suggesting a move toward a structured, creator-focused platform.

🗞 #aitestkitchen

Comet enters early testing as Perplexity debuts its agentic browser

Perplexity's Comet Browser integrates its AI agent directly into core browsing functions, enabling autonomous tab management, task execution, and contextual content understanding—positioning it as a full-stack assistant for advanced workflows.

🗞 #perplexity

Claude Sonnet 4 and Opus 4 spotted in early testing round

Anthropic is nearing launch for its Claude 4 model family, currently in limited internal testing. Claude Sonnet 4 and Claude Opus 4 are under evaluation with strict access limits and higher safety classification due to advanced capabilities.

🗞 #claude

Devstral from Mistral AI tops open-source benchmarks for agentic tasks

Mistral AI and All Hands AI launched Devstral, a 24B-parameter open-source language model for software engineering, outperforming key benchmarks. It runs locally, supports 128K tokens, and is available under Apache 2.0 via multiple platforms.

🗞 #mistral

Opus 4 outperforms GPT-4.1 and Gemini 2.5 Pro in coding benchmarks

Anthropic introduced Claude Opus 4 and Sonnet 4, advanced hybrid AI models optimized for coding, reasoning, and long agent workflows. Available via API and cloud platforms, both models show strong benchmark results and improved session performance.

🗞 #claude

Claude Sonnet 4 now available to all Perplexity Pro users

Perplexity Pro now includes access to Claude Sonnet 4 and Sonnet 4 Thinking, offering upgraded reasoning and coding functions with an optional step-by-step mode for complex tasks. Available on both web and mobile platforms.

🗞 #perplexity

Perplexity prepares Scheduled Tasks feature to automate your web browsing

Perplexity is developing a Scheduled Tasks feature to automate routines like news digests or content publishing, optimized for its Comet browser, which supports deeper web automation across platforms such as LinkedIn and Ghost.

🗞 #perplexity

xAI prepares screen-sharing for Grok voice mode on iOS

xAI is preparing screen-sharing for Grok’s voice mode on iOS, allowing real-time screen access for tasks like translation and app guidance. Android users trail behind as Grok 3.5 development progresses with no confirmed release date.

🗞 #grok

AI Studio now supports vibe coding with Gemini Code Assist

Google AI Studio now supports in-app coding with real-time feedback and code diffs. It accesses live web content via URLs and adds the compact, multimodal Gemma 3n E4B model, expanding its coding and contextual processing capabilities.

🗞 #aistudio

xAI preps Workspace sharing as Grok 3.5 model timeline shifts

xAI is developing Workspace sharing features, hinting at future support for reusable, public templates. Meanwhile, Grok 3.5's release appears delayed despite backend references, suggesting a cautious rollout for integrated updates.

🗞 #grok

Mistral AI rolls out faster, more powerful Agents inside Le Chat

Mistral AI is replacing its Agent Builder with a new Agents feature in Le Chat, offering faster response times, advanced customization, tool access, and organizational sharing, all accessible even to free-tier users.

🗞 #mistral

Google developing speed-focused Veo 3 variant spotted in Flow Editor

Google is testing a "Fast" version of its Veo 3 video generation model, aimed at quicker, lower-cost output with potential quality and feature trade-offs, including possible absence of audio, to meet varying production needs.

🗞 #aitestkitchen

NotebookLM set to support public notebooks and video summaries

Google is preparing public sharing and multimedia summary features for NotebookLM, allowing content to reach wider audiences without registration. These updates align with its focus on research and content repurposing use cases.

🗞 #notebooklm

Google experiments with computer use environments in AI Studio

Google is expanding AI Studio with a feature called "computer use," enabling browser or OS-level task simulation in sandboxed environments. Though limited now, it signals a move toward applied AI testing and autonomous agent development.

🗞 #aistudio