Anthropic tests Claude Code upgrade to rival Codex Superapp Anthropic’s “Epitaxy” revamps Claude Code’s desktop app with Plan, Tasks, Diffs, code preview, multi-repo support, custom agents, and a local Coordinator Mode, signaling a shift toward power-user developer workflows.

Anthropic tests Claude Code upgrade to rival Codex Superapp

xAI prepares credits system for upcoming Grok Build launch xAI is preparing a credits-based pricing model for Grok Build, its upcoming coding platform with local CLI and web access. Features like Model Arena and bundled monthly credits point to a commercial launch aimed at competing with Codex and Claude Code.

xAI prepares credits system for upcoming Grok Build launch

OpenAI develops unified Codex app and new Scratchpad feature OpenAI is developing a Scratchpad feature for Codex to run parallel tasks and moving toward a unified desktop app that could merge ChatGPT and Atlas. Code hints at managed agents with persistent workflows, alongside speculation of a GPT-5.5 release.

OpenAI develops unified Codex app and new Scratchpad feature

Notion tests Computer to expand capabilities of its AI offering Notion is expanding its agent strategy by introducing a “Computer” capability and integrating Anthropic’s Claude Managed Agents. The move enables teams to assign autonomous, multi-step tasks from Notion, with agents operating beyond the workspace via managed infrastructure.

Notion tests Computer to expand capabilities of its AI offering

Google prepares broader rollout of Skills for Gemini and AI Studio Google is preparing to expand Gemini “Skills” to AI Studio and consumer versions, enabling reusable instruction sets and scheduled tasks. The move would support automated workflows, standardize outputs across projects, and align Gemini with competing AI platforms ahead of I/O 2026.

Google prepares broader rollout of Skills for Gemini and AI Studio

Anthropic develops managed 24/7 Agent for Claude mobile apps Anthropic appears to be preparing a consumer version of its managed “Conway” agents within the Claude iOS app, signaling a push toward unified, persistent agent functionality across mobile, web, and desktop amid rising competition in autonomous AI tools.

Anthropic develops managed 24/7 Agent for Claude mobile apps

Anthropic launches advisor tool for Claude Platform API users Anthropic has introduced the advisor tool on the Claude Platform, enabling developers to pair Opus as an advisor with Sonnet or Haiku as executors. This setup delivers near-Opus reasoning at lower cost, boosting benchmark performance while maintaining efficient agent operations.

Anthropic launches advisor tool for Claude Platform API users

Perplexity launches Personal Finance powered by Plaid Perplexity expands its Plaid integration from portfolio tracking to a full personal finance dashboard, linking bank accounts, credit cards, loans, and investments. Users can analyze spending, monitor debt, and calculate net worth through AI-driven insights.

Perplexity launches Personal Finance powered by Plaid

Anthropic launches Claude Cowork in General Availability Anthropic has made Claude Cowork generally available on all paid plans, adding role-based access controls, group spend limits, usage analytics, expanded OpenTelemetry, and a Zoom connector to support governance, compliance, and cross-team workflow management.

Anthropic launches Claude Cowork in General Availability

Google adds NotebookLM directly into Gemini for paid subscribers Google is adding Notebooks to Gemini, creating a shared workspace that syncs with NotebookLM. Users can group chats, files, and instructions in one project, then switch between tools for research and outputs, as Google unifies both products under Gemini.

Google adds NotebookLM directly into Gemini for paid subscribers

Anthropic launches Claude Managed Agents for businesses Anthropic has launched Claude Managed Agents in public beta, offering composable APIs and production-grade infrastructure for building and deploying cloud-hosted AI agents. The suite supports secure execution, orchestration, and multi-agent workflows at scale.

Anthropic launches Claude Managed Agents for businesses

Meta unveils Muse Spark AI model and new Contemplating mode Meta has launched Muse Spark, a new AI model built on a rebuilt pretraining stack with upgraded architecture, optimization, and data pipelines. It matches prior performance using over 10x less compute and ranks 4th on Artificial Analysis, targeting scalable research use.

Meta unveils Muse Spark AI model and new Contemplating mode

Zhipu AI launches open-source GLM-5.1 model for coding tasks Z.ai has launched GLM-5.1, a flagship model built for agentic engineering and long-horizon coding, capable of running up to eight hours on a single task. It leads open-source benchmarks like SWE-Bench Pro and targets developers with deep tool and repo support.

Zhipu AI launches open-source GLM-5.1 model for coding tasks

Anthropic announces Claude Mythos for cybersecurity research Anthropic introduced Claude Mythos Preview, an AI model that autonomously detects and exploits zero-day vulnerabilities. It has uncovered critical flaws across major systems and is available to select partners, with $100 million in credits supporting cybersecurity efforts.

Anthropic announces Claude Mythos for cybersecurity research

Anthropic likely tests managed 24/7 AI agents for Businesses Anthropic’s Conway project is evolving into an enterprise-focused agent platform with extensions, browser control, and cryptographically verified webhooks. Its design supports secure system-to-system automation, positioning Conway as managed infrastructure for teams building workflows around Claude.

Anthropic likely tests managed 24/7 AI agents for Businesses

Telegram adds AI text editor and upgraded polls for all users Telegram has launched an update featuring a Cocoon AI-powered Text Editor that corrects, translates, and restyles messages with privacy protection and no data retention. The release also upgrades polls with media support, custom options, quiz modes, and advanced admin controls.

Telegram adds AI text editor and upgraded polls for all users

Atomic Bot now runs local AI models on your computer Atomic Bot now runs fully offline via Ollama, letting users deploy OpenClaw on local hardware with no API keys or cloud reliance. It supports major open models, 40,000+ skills, and multi-messenger automation, packaged as a native macOS and Windows app.

Atomic Bot now runs local AI models on your computer

Google prepares Jules V2 agent capable of taking bigger tasks Google is reportedly developing “Jitro,” a next-gen Jules coding agent centered on goal-driven development. Instead of task-based prompts, it would pursue KPI-level objectives within a dedicated workspace, signaling a shift toward autonomous, outcome-focused AI collaboration for engineering teams.

Google prepares Jules V2 agent capable of taking bigger tasks

OpenAI tests next-gen Image V2 model on ChatGPT and LM Arena OpenAI is testing Image V2 on LM Arena in three variants, showing stronger UI rendering, accurate text, and prompt fidelity. Some users access it via A/B tests. A launch could counter Google’s lead, though pricing and quality retention remain uncertain.

OpenAI tests next-gen Image V2 model on ChatGPT and LM Arena

Microsoft releases 3 new models on MAI Playground Microsoft has added MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 to Foundry, opening its media AI stack to developers beyond Copilot. The models support speech-to-text, custom voice generation, and image creation, with enterprise pricing and broader deployment.

Microsoft releases 3 new models on MAI Playground

Cursor launches Cursor 3 with Agents Window and Design Mode Cursor 3 launches with parallel agent execution, agent tabs, and a redesigned interface supporting local, SSH, worktrees, and cloud setups. It adds native worktree management, new commands, faster diffs, and enterprise security controls for scalable AI-driven development.

Cursor launches Cursor 3 with Agents Window and Design Mode

Google debuts Gemma 4 open AI models for local use Gemma 4 is Google’s latest open model family under Apache 2.0, built for on-device AI across edge, mobile, and high-performance systems. With up to 256K context, multilingual support, and native tool use, it enables agentic workflows, multimodal tasks, and complex reasoning.

Google debuts Gemma 4 open AI models for local use

Meta tests Paricado model family, Health, and Document agents Hands-on testing indicates Meta’s Avocado models are further along than reported, with multiple variants, including a reasoning build, already active. A new Paricado family and specialized agents also suggest broader AI development despite launch delays.

Meta tests Paricado model family, Health, and Document agents

Exclusive: Anthropic tests its own always-on "Conway" agent Anthropic is developing Conway, a standalone Claude environment with extensions, webhooks, Chrome connectivity, and always-on agent capabilities. The tool signals a shift toward persistent, extensible workflows and deeper third-party integration beyond the current desktop chat model.

Exclusive: Anthropic tests its own always-on "Conway" agent

Anthropic tests new Claude Code desktop UI amid source code leak Anthropic appears to be testing Epitaxy, a possible new interface for Claude Code Desktop featuring expanded hotkeys and playful visuals, as a source map leak exposed internal models like Capybara and Strudel, fueling speculation about its roadmap.

Anthropic tests new Claude Code desktop UI amid source code leak

Google prepares 3D Avatars for Gemini and Remy tools for learning Recent Gemini builds reveal an Avatar option for creating 3D portraits, a student-focused Remy learning mode, and signs of Skill Support reaching some Ultra users. With Google I/O nearing, these additions suggest upcoming launches or announcements.

Google prepares 3D Avatars for Gemini and Remy tools for learning

Holo Company launches Holo3, SOTA Computer Use model Holo Company has introduced Holo3, an enterprise automation model scoring 78.85% on OSWorld-Verified while operating at one-tenth the cost of larger rivals. With open weights, API access, and agentic training, it supports complex, multi-application business workflows.

Holo Company launches Holo3, SOTA Computer Use model

Meta unveils prescription-first AI glasses and new AI features Meta has introduced prescription-ready Ray-Ban Meta Blayzer and Scriber Gen 2 glasses, targeting everyday wearers with lighter, customizable frames starting at $499. The update focuses on accessibility and adoption, alongside new AI features and expanded US availability.

Meta unveils prescription-first AI glasses and new AI features

Base44 launches skills library for Superagents with 130+ options Base44 has launched a Skills library for Superagents, offering 130+ domain-specific capabilities across seven categories. Skills persist across sessions, stack within workflows, and can be custom-built in chat, giving users durable, reusable agent expertise from day one.

Base44 launches skills library for Superagents with 130+ options

Google launches Veo 3.1 Lite at half price of Veo 3.1 Fast Google has introduced Veo 3.1 Lite, a lower-cost video generation model available via the Gemini API and Google AI Studio. Supporting text-to-video and image-to-video at 720p and 1080p, it offers scalable pricing, undercutting Veo 3.1 Fast as Google also reduces Fast’s price.

Google launches Veo 3.1 Lite at half price of Veo 3.1 Fast