Developer AI news
April 2026
Anthropic Launched Claude Managed Agents in Public Beta
Anthropic opened a public beta for Claude Managed Agents, a new platform that simplifies agent deployment by handling backend infrastructure, allowing developers to define tasks, tools, and guardrails.
Guide: Build Automated Video Ads with ElevenLabs Flows
A guide on using ElevenLabs Flows, a new workflow builder, to transform product photos into finished video ads by bundling image, video, voice, and music generation in one place.
OpenAI Launches Pro Tier with 5x Codex Usage for Agentic Coding
OpenAI introduced a new Pro tier at $100/month, offering five times the Codex usage compared to the Plus plan, specifically designed to support heavy agentic coding workflows.
Spacelift Intelligence: AI Infrastructure Suite for Platform Teams
Spacelift Intelligence is a new AI infrastructure suite designed to help platform teams ship infrastructure as fast as developers code, streamlining the delivery of AI-related infrastructure.
Guide: Automate Business Workflows with Custom Notion Agents
A step-by-step guide on building Notion Custom Agents that run a company's recurring work on a schedule, automating tasks like inbound leads, campaigns, and accounting without adding complexity.
Intel Joins Elon Musk's Terafab Project for AI Chip Fabrication
Intel has partnered with Elon Musk's Terafab project, alongside SpaceX and Tesla, to design and fabricate chips for advanced AI applications including robotaxis, Optimus robots, and space technology, aiming to produce 1 TW/year of compute.
Mystery Model HappyHorse-1.0 Tops Video Leaderboard
An unknown model, HappyHorse-1.0, has emerged at the top of Artificial Analysis' video leaderboard, outperforming ByteDance's Seedance 2.0, with its origin and company behind it currently undisclosed.
Clicky: AI-Powered On-Screen Teaching Tool
Clicky is an on-screen teaching tool that uses Claude for reasoning and ElevenLabs for voice to provide real-time, interactive guidance by highlighting exactly where to click to learn new software.
Z AI's GLM-5.1 Open-Source Model Tops Coding Leaderboard
Z AI released GLM-5.1, an open-source coding model that achieved a 58.4 on SWE-Bench Pro, surpassing GPT-5.4 and Claude Opus 4.6, and demonstrating sustained performance on long-horizon autonomous coding tasks for up to 8 hours.
Anthropic Unveils Project Glasswing with Unreleased Claude Mythos Preview
Anthropic introduced Project Glasswing, a cybersecurity coalition built around Claude Mythos Preview, a powerful frontier model capable of flagging thousands of security vulnerabilities, deemed too dangerous for public release and restricted to defensive security applications.
HeyGen Releases Avatar V, Eliminating AI Identity Drift
HeyGen's Avatar V model creates realistic video avatars from 15-second phone recordings, claiming to eliminate identity drift and allowing users to swap outfits and backgrounds without re-filming.
Wispr Flow: Elite Voice Recognition for All Platforms
Wispr Flow offers highly accurate and fast voice recognition with intelligent text replacements, available for Mac, PC, iPhone, and Android, suitable for integration into various applications.
xAI Training Seven New Models, Including 10-Trillion Parameter System
xAI is simultaneously training seven new models on its Colossus 2 supercomputer, including systems up to 10 trillion parameters, indicating aggressive architectural experimentation to compress the timeline for finding effective approaches.
Meta Superintelligence Labs Ships First Model: Muse Spark
Meta's Muse Spark, a multimodal reasoning model from Alexandr Wang's Superintelligence Labs, handles voice, text, and image inputs with competitive benchmarks against frontier rivals like Opus 4.6 and GPT 5.4 on reasoning. It is particularly strong in health reasoning and is proprietary, with API access for selected partners.
OpenAI Develops New Cybersecurity Model
OpenAI has reportedly built a new model with advanced cybersecurity capabilities, akin to Anthropic's Mythos, with plans for release to a select group of partners.
Anthropic's Claude Cowork Now Generally Available
Claude Cowork, Anthropic's agentic system, has moved from research preview to general availability for all paid plans, offering enhanced agentic capabilities for users.
Meow: Infrastructure for AI Agents to Manage Money
Meow provides financial rails, enabling AI agents to open bank accounts, issue payment cards, and manage money, simplifying financial operations for agent builders.
Teleport Introduces Beams for Secure Agentic AI Infrastructure
Teleport launched Beams, a new solution enabling users to run AI agents securely across infrastructure with built-in identity, control, and trusted runtimes. This enhances the security and manageability of agentic AI deployments for developers.
H Company Releases Holo 3 Open-Weight, State-of-the-Art Computer-Use Agent
H Company launched Holo 3, a new open-weight, state-of-the-art computer-use agent. This model is designed for advanced autonomous interaction with computer interfaces, offering significant capabilities for agentic AI development.
Fauna: Creative Agent for AI Model Routing and Prompting
Fauna is a built-in creative agent designed to manage AI model routing and prompting. It streamlines the process of interacting with various AI models, offering developers a more efficient way to orchestrate creative AI workflows.
Bland AI Launches Norm, a Voice AI Assistant for Prompt-Based Phone Agent Creation
Bland AI introduced Norm, a voice AI assistant that enables users to create fully functional phone agents from a single prompt, eliminating the need for extensive development work. Norm automates the generation of agent logic, conditions, and integrations like calendar booking.
Marksnip Chrome Extension Enhances AI Coding Context with Markdown Documentation
Marksnip is a Chrome extension that allows developers to quickly convert web documentation into markdown files, which can then be used to provide rich context to AI coding agents. This improves the accuracy and relevance of AI-generated code by feeding agents up-to-date project-specific information.
Z AI Releases GLM-5V-Turbo 'Vision Coding' Model for Code Generation from Visuals
Z AI introduced GLM-5V-Turbo, a new 'vision coding' model capable of generating runnable code directly from visual inputs like screenshots, design drafts, and interfaces. This model streamlines the development process by translating visual concepts into functional code.
Contra Labs Launches AI Creative Tool Evaluation Platform with Leaderboards and Benchmarks
Contra Labs emerged from stealth to introduce a new evaluation platform for AI creative tools, featuring leaderboards, datasets, and benchmarks. The platform focuses on human creative taste, providing developers with metrics to assess and improve AI-generated content.
Alibaba Launches Wan2.7-Image for Unified Image Generation, Editing, and Text Rendering
Alibaba released Wan2.7-Image, a new unified image model capable of generating, editing, and rendering text across 12 languages. It supports up to 12 consistent images per prompt, enhancing creative production workflows.
Liquid AI Releases LFM2.5-350M Small Model for Tool Use and On-Device Agents
Liquid AI introduced LFM2.5-350M, a small open model designed for efficient tool use and on-device agent deployment. It reportedly outperforms models twice its size, making it suitable for resource-constrained consumer devices.
Strands Agents: Innovation AI for Production Backends and Code Generation
Strands Agents provides innovation AI solutions for developers, covering production backends, physical robots, and code generation. This platform aims to simplify complex AI implementations and enable self-writing code.
Sakana AI Opens Beta for Marlin Autonomous AI Research Assistant
Japanese AI startup Sakana AI launched beta testing for Marlin, an autonomous AI research assistant capable of working up to 8 hours straight on business-related tasks. This tool aims to streamline research workflows for developers and businesses.
Merge Gateway for Production AI Routing, Cost Controls, and Observability
Merge Gateway offers a solution for shipping production AI faster by providing built-in routing, cost controls, and observability features. This tool helps developers manage and optimize their AI deployments efficiently.
Salesforce Upgrades Slackbot with 30 New Agent Capabilities
Salesforce enhanced its Slackbot agent with 30 new capabilities, including reusable skills, MCP connections, and desktop operation. This update significantly expands the automation and integration possibilities for developers building on the Slack platform.
PrismML Launches Bonsai, a 1-bit Compressed Open-Source AI Model for Consumer Hardware
PrismML unveiled Bonsai, a tiny open-source AI model utilizing 1-bit compression, enabling it to run efficiently on consumer hardware without performance degradation. This breakthrough allows for powerful AI capabilities on resource-constrained devices.
Google Releases Veo 3.1 Lite, a Budget Video Generation Model for Developers
Google introduced Veo 3.1 Lite, a new cost-efficient video generation model designed for developers. It offers video clip generation up to 8 seconds at half the cost of its 'Fast' variant, making advanced video creation more accessible.
Anthropic Claude Code Source Code Leaked via npm, Revealing Internal Architecture
Anthropic accidentally exposed Claude Code's full TypeScript codebase (512,000+ lines) via an npm source map leak, detailing its three-layer memory architecture, persistent KAIROS agent, ULTRAPLAN deep-planning system, and multi-agent coordination. The leak also highlighted a potential supply chain attack on the axios npm package, urging users to migrate to the native installer and rotate API keys.
Arcee AI Releases Trinity-Large-Thinking Open-Weight Reasoning Model
Arcee AI introduced Trinity-Large-Thinking, an open-weight reasoning model that rivals Opus 4.6 on agent benchmarks while operating at significantly lower cost. This model offers a cost-effective solution for long-horizon agentic tasks.
Verdent AI Coding Workspace Routes Tasks to Multiple Frontier Models
Verdent is an AI coding workspace that streamlines development from idea to code, breaking features into steps and building implementations across multiple AI models in parallel. It intelligently routes tasks to different frontier models like Claude, GPT-5, and Codex based on task requirements.
OpenAI's 'Project Stagecraft' Uses Freelancers for Occupation-Specific Model Training Data
OpenAI's 'Project Stagecraft' employs up to 4,000 freelancers to generate occupation-specific training data, focusing on knowledge work. This initiative aims to train models with real-world expertise by simulating professional workflows and providing context, goals, and deliverables.
Cursor 3 Released with Multi-Workspace Support for Coding Agents
Cursor 3 features a full interface rebuild, introducing multi-workspace support that enables developers to run local and cloud coding agents in parallel across multiple repositories. This update significantly enhances developer productivity and agent management.
Microsoft MAI-Transcribe-1 Speech-to-Text Model in Public Preview
Microsoft launched MAI-Transcribe-1, a new speech-to-text model now available in public preview. It achieves top accuracy benchmarks across 25 languages, offering enhanced transcription capabilities for developers.
Alibaba Releases Qwen3.6-Plus Reasoning Model with 1M-Token Context
Alibaba introduced Qwen3.6-Plus, a new reasoning model that competes with Opus 4.5 on coding agent benchmarks. It features native 1M-token context and multimodal input capabilities, enhancing its utility for complex agentic tasks.
ByteDance Seedance 2.0 AI Video Generator Broadly Available
ByteDance's Seedance 2.0, an AI video generator, is now widely accessible and has achieved top rankings on Artificial Analysis' video leaderboards. This release provides developers with a powerful tool for high-quality video content creation.
Google DeepMind Releases Gemma 4 Open Models with Apache 2.0 License
Google DeepMind launched Gemma 4, a family of four open models (2B, 4B, 26B MoE, 31B) featuring vision and multi-step agent capabilities. Released under an Apache 2.0 license, these models offer commercial flexibility and include smaller variants capable of offline operation on mobile devices.
March 2026
Claude Code Introduces New Auto Mode for Hands-Free Coding
Claude Code has rolled out a new auto mode, providing a hands-free coding permission system. This enhancement streamlines the development workflow, allowing developers to delegate more coding tasks to Claude with automated permissions.
Anthropic's Claude Gains Remote Computer Control with Dispatch
Dispatch allows Claude to remotely perform tasks on a user's PC from their phone, enabling file access, organization, and other operations while keeping conversations synced across devices. This feature requires updates to both Claude Desktop and mobile apps, enhancing agentic capabilities.
Sierra Introduces Ghostwriter, an AI Agent for Building AI Agents
Bret Taylor’s Sierra launched Ghostwriter, an innovative AI agent designed to build other AI agents. This tool enables companies to rapidly create specialized customer service bots capable of handling interactions across voice, chat, and over 30 languages.
Ai2 Releases MolmoWeb, an Open-Source Web Browsing Agent
The Allen Institute for AI (Ai2) has released MolmoWeb, an open-source web browsing agent. This tool provides developers with a foundation for building and experimenting with AI agents capable of navigating and interacting with web content.
Slack Integrates AI Agents with Agentforce for Enhanced Workflows
Agentforce brings powerful AI agents directly into Slack, allowing users to interact with them via DMs or @mentions to pull Salesforce insights, update records, and create canvases. This integration offers ready-made templates and custom agent building, streamlining agentic workflows within the communication platform.
Alibaba.com Introduces Accio Work for Qwen-Powered AI Agent Teams
Alibaba.com's Accio Work enables businesses to deploy Qwen-powered AI agent teams that handle complex, multi-step tasks across various functions without code. Users can configure agents, combine teams, and encapsulate domain expertise into reusable skills, shifting focus from task management to defining outcomes.
ARC-AGI-3 Benchmark Challenges Frontier AI Models, Scores Below 1%
François Chollet's ARC Prize Foundation released ARC-AGI-3, a new interactive reasoning benchmark where top AI models score below 1% while humans achieve 100%. This version challenges agents to discover rules, form goals, and plan strategies from scratch without instructions, pushing the boundaries of genuine AI reasoning.
Google Research Introduces TurboQuant for 6x LLM Memory Compression
Google Research published TurboQuant, a compression algorithm that shrinks LLM cache memory by over 6x and boosts processing speed by up to 8x on Nvidia H100 chips, all with virtually zero accuracy loss. This technique compresses conversation logs without requiring model retraining.
Mistral Ships Voxtral TTS for Lightweight, Multilingual Voice Cloning
Mistral released Voxtral TTS, a lightweight voice AI model capable of cloning any speaker's voice from a 3-second audio clip and generating natural-sounding speech across 9 languages. This offers significant capabilities for multilingual speech agents.