OpenAI Announces GPT-5.4 with Native Computer Control
OpenAI has officially launched GPT-5.4, a major update that integrates reasoning, coding, and agentic capabilities into a single model. The most significant news with the GPT-5.4 release is its ability to exercise native computer control through Playwright, allowing it to perform complex, multi-step tasks directly on a computer.
The model boasts a 1 million token context window and shows impressive performance on industry benchmarks. It achieved 87.3% accuracy on banking analyst spreadsheet tasks and 75% on the OSWorld benchmark, surpassing the human baseline. However, its capabilities also earned it a "high" cybersecurity risk rating after it scored 88% on professional hacking challenges.
Alongside the new model, OpenAI also released a native Codex Windows app and a suite of financial services tools designed to leverage its enhanced analytical power.
The Cinematic AI Video Race: Kling 3.0 and Seedance 2.0
The field of AI video generation saw major advancements with two key players making waves. These tools are moving beyond short, distorted clips to produce more polished, cinematic content.
Kling 3.0 Goes Global
Kling 3.0 has been rolled out worldwide, offering native 1080p cinematic video generation for clips up to 30 seconds long. Its node-based editing canvas allows for easy actor swaps and motion transfers. The model has quickly climbed to the top spot on the Artificial Analysis Text-to-Video leaderboard, ahead of competitors like Runway and Veo, and offers a free tier for users. Check out the full list of features and our technical analysis of Kling AI.
Seedance 2.0's Viral Rise
ByteDance's AI video model, Seedance 2.0, gained viral attention for its ability to create realistic videos from a combination of text, images, and audio. It excels at maintaining character consistency and producing believable physics. Users can access a few free daily generations through the Dreamina website without needing a VPN.
Rise of Specialized AI Agents and Models
Beyond general-purpose models, a new wave of specialized AI agents is emerging to tackle specific, complex workflows. These tools focus on everything from software development to web browsing and film production.
Tool Name | Primary Function | Key Feature |
|---|---|---|
Rovo Dev | Software Development | Reduces PR cycle time by 45% and auto-resolves security vulnerabilities. |
Luma Creative Agents | Film Production | Spawns an autonomous AI film crew to iterate on lighting and camera moves. |
Hyperbrowser | Web Automation | Completes multi-step tasks like booking flights with 94% success, powered by GPT-5.4. |
Liquid AI LocalCowork | Desktop Agent | Runs a fully local agent on a MacBook with zero network calls. |
New AI Models and Infrastructure Updates
The industry's foundation is also strengthening with more efficient models and faster processing technologies. Alibaba's Qwen3.5-9B model is a standout, outperforming much larger models like OpenAI's gpt-oss-120B while requiring only 6GB of RAM. This efficiency allows developers to run it for free on standard hardware like a MacBook Air.
Under the hood, technologies like FlashAttention-4 are making every chatbot faster. Developed by Together AI and Tri Dao, this latest version achieves up to 2.7 times the speed of previous methods on Blackwell GPUs, an upgrade that benefits nearly every AI model in use today.
Productivity Tools Get an AI Upgrade
AI is also being integrated more deeply into everyday productivity software. Claude Code Desktop introduced local scheduled tasks, allowing users to run recurring jobs on their machines. For spreadsheet users, ChatGPT for Excel can now build, analyze, and fix spreadsheets using natural language commands.
Google has rolled out Google Canvas to all US users, providing a side panel in Search to generate documents and dashboards from prompts. Additionally, Willow offers a speech-to-text service that is three times more accurate than built-in dictation, with context-aware formatting and grammar correction. Find more daily tool updates in our news section.
