Runway Launches AI Creative Agent, Runway Agent
Runway has introduced a new AI creative agent that can generate full videos — from concept to final delivery — entirely through conversation.
Meet Runway Agent, a new AI creative partner that helps you ideate and execute fully finished, sound designed and edited videos. All with just a simple conversation. From ads to shorts to content for social, Runway Agent makes it easy to produce more of what you need. The platform marks a significant leap in AI-assisted video production, shifting from single-shot generation to complete end-to-end creative workflows driven by natural language. Users describe the output they want, and the agent handles storyboarding, editing, sound design, and final rendering autonomously.
Claude Code Weekly Limit Raised by 50%
Anthropic announced that effective immediately through July 13, the weekly usage limit for Claude Code is increased by 50% for Pro, Max, Team, and seat-based enterprise users. The change reflects surging demand for AI-assisted coding in professional development environments.
Cursor Introduces Cloud Agent Running in Pre‑configured Environments
Cursor added a cloud agent feature that runs agents inside ready‑to‑use development environments — cloned repos, installed dependencies, and toolchain credentials included. Teams configure them the same way they would set up a laptop for an engineer, then let agents operate autonomously in the cloud.
Google Unveils First Gemini‑Powered Laptop, Googlebook
Google introduced the Googlebook, the first laptop designed specifically for Gemini Intelligence, crafted for heavyweight AI performance.
The device positions Gemini Intelligence as the core operating layer rather than a bolt-on feature. Built with on-device AI acceleration, it promises to run large models locally with low latency. The announcement signals Google's intent to compete directly in the AI-first hardware space, bringing Gemini capabilities natively to the laptop form factor.
Sam Altman: Codex Is the Best AI Coding Product
Altman called Codex the best AI programming product and announced two months of free usage for enterprises that switch within 30 days. "Codex is the best AI coding product and we want to make it easy to try," he wrote. The promotion is designed to accelerate enterprise adoption of Codex as the competitive landscape for AI coding tools intensifies.
Codex Enterprise Promotion: Two Free Months for Switching
OpenAI launched a Codex enterprise migration offer: new enterprise customers who switch within 30 days receive two months of free Codex usage. Eligible enterprise customers can bring their entire team to Codex and get two free months for new users. The promotion targets teams currently using competing coding assistants.
Claude Paid Plans to Offer Monthly Coding‑Only Credits
Starting June 15, paying Claude users will receive a monthly dedicated credit for programmatic usage. The credit covers usage of the Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps built on the Agent SDK. This gives developers predictable monthly allocation for agentic coding workflows without consuming their general chat quota.
Qwen3.6-Plus Arrives on Nous Portal with Limited Free Access
Qwen3.6-Plus is now available on Nous Portal for a limited free period, alongside the newly introduced Hermes Agent. The model from Alibaba's Qwen team represents continued progress in open-weight frontier models, joining the growing ecosystem of accessible large language models available to developers and researchers.
Snap Shifts 10PB+ Daily Data to GPU Pipelines, Cuts Costs 76%
Snap migrated over 10 petabytes of daily data processing to GPU‑accelerated pipelines on Google Cloud, reducing job costs by 76%. The migration was led by Prudhvi Vatala, Head of Engineering Platforms, who detailed how the team moved A/B testing infrastructure serving nearly a billion users onto GPU-accelerated pipelines.
NVIDIA and SAP Partner to Build Enterprise AI Agents
NVIDIA and SAP are integrating specialized AI agents into the SAP Business AI platform, delivering security, governance, and execution controls built into the platform. NVIDIA OpenShell is helping teams move from agent development to trusted production deployment across enterprise environments.
NVIDIA and IneffableLabs Build Infrastructure for Large‑Scale RL Agents
NVIDIA is collaborating with IneffableLabs to design training infrastructure for massive reinforcement learning agents, aiming to accelerate discovery across science and industry. Engineers from both teams are exploring how to create training pipelines that will allow agents to discover new knowledge autonomously.
I get some anxiety not using the smartest-available model. But sometimes I don't mind if it's really slow. I wonder if we should focus more on a price/speed tradeoff relative to a price/intelligence tradeoff.
Sam Altman
Claude Code Adds /goal Command to Keep Tasks on Track
Claude Code introduced a /goal feature that helps Claude stay focused and work continuously until a target task is completed. The command is part of a broader effort to keep AI coding agents working until the job is done, addressing one of the most common pain points in autonomous coding workflows.
Perplexity Emphasizes Default Security: Hardware‑Isolated Sandboxes
Perplexity stated its Computer product is secure by default; each task runs in a dedicated hardware-isolated sandbox with VPC-level storage and compute separation. Agents are authenticated with short-lived proxy tokens instead of raw API keys, reducing the surface area for credential leaks.
PayPal Runs 74,000 Tasks Weekly on Perplexity Enterprise
Perplexity published a case study showing PayPal uses its enterprise edition for model validation, channel performance, market trend research, competitive intelligence, and product analysis — executing 74,000 tasks per week. The deployment demonstrates how large enterprises are operationalizing AI research platforms at scale.
Replit Hosts Vibecon: Code‑Driven Creative Installations in NYC
Replit will hold Vibecon in New York on June 17–18, themed "Code is the new creativity," transforming NYC into a living gallery of code-driven AI installations. Speakers include director Spike Jonze, media artist Refik Anadol, and digital artist Gmunk, alongside Replit CEO Amjad Masad.
Luma Creative Agent: From Concept to Marketing Assets Automatically
Luma published a blog introducing Luma Agents; after a user uploads a design and applies it to a product, the agent automatically generates every promotional image — from concept to campaign-ready in minutes. The agents serve as force multipliers, handling planning, generation, iteration, and refinement across the full creative workflow.
Kling AI to Explore AI Cinema at 2026 Cannes Film Festival
Kling AI will hold a dedicated session at the 2026 Cannes Film Festival, as part of the Cannes Conference at the Marché du Film, exploring the current landscape and future evolution of AI-supported filmmaking. The session covers the journey from creative possibility to production reality.
Hunyuan Hy3 Preview Now Available
The preview version of the Hunyuan Hy3 model is live on the gmi_cloud platform.
Long‑Horizon Agent Leaderboard Preview Published
The Kimi team integrated FrontierCS into Harbor and released a preview leaderboard for long‑horizon agents, supporting up to 835 evaluation rounds.
Opus 4.7 Fast Mode Preview Opens on Multiple Platforms
A research preview of Claude Opus 4.7 fast mode is now available on Cursor, Emergent Labs, Factory, v0, and Warp.
Voice Model Dramabox Released
ResembleAI's research team launched Dramabox, a new voice AI model designed for cinematic‑grade performance.
Diffusers 0.38.0 Released with New Pipelines
Hugging Face launched Diffusers 0.38.0, shipping new pipelines and multiple library improvements.
GLM Models Arrive on Tensorix Platform
Z.ai partnered with Tensorix to bring GLM models to the platform, offering developers access to efficient cutting‑edge models.
2,000 Developer Inquiries in Three Hours
Agents Can Now Train Models; GLM 5.1 Leads
Run a 24/7 AI Researcher on a Laptop for Free
AI Supercomputer via Browser or Telegram
Samsung Exec Uses AI Clone for Cross‑Country Comms
Step 3.5 Flash Generates Playable Game in Minutes
Vidu Q3 Suite Upgraded: Faster and More Efficient
The Vidu Q3 suite received an upgrade optimized for storytelling and creative scenarios, with a free trial code available.
Vercel AI Gateway Reveals Real‑World AI and Agent Usage
Vercel's AI Gateway data shows Google leading in production scale, Anthropic dominating in coding spend, and OpenAI growing fast since GPT-5.4. Open‑source models continue gaining ground across production workloads.
Google Launches Gemini Intelligence Across Devices
Google introduced Gemini Intelligence, bringing the best of Gemini to advanced devices including Samsung Galaxy and Pixel phones, with broader Android roll-out planned.
Apple: On‑Policy Distillation — Where It Helps, Where It Hurts
Apple researchers published a study examining when on‑policy distillation improves model performance and when it backfires, offering practical guidance for training pipelines.
SenseNova-U1: Unifying Multimodal Understanding and Generation
A new NEO-unify architecture combines multimodal understanding and generation in a single model framework, targeting more integrated AI perception systems.
UK AI Safety Institute Assesses Mythos Cyber Capabilities
The UK's state AI Security Institute found Mythos represents a significant gain in cyber capabilities, but pointed out that similar models are being developed at Google and OpenAI.
The Increasing Levels of AI Agent Autonomy
From /skill (preset prompts) to /plan (human-refined inputs) to /goal (AI-evaluated outcomes), the spectrum of agent autonomy is expanding rapidly in coding tools.
Luma Agents Build Full Product Launch Asset Pipelines
Luma Labs showed how its AI agents automatically generate teaser countdown assets and launch marketing materials, from first hint to final reveal.
Hugging Face CEO Urges US AI Community to Support Open Source
Clement Delangue called on the American AI community to recognize the value of international open‑source AI, pointing to Chinese models like DeepSeek and Qwen.