Alongside the GPT-5.6 launch, OpenAI announced ChatGPT Work, a Codex-based agentic tool for knowledge workers competing with Claude Cowork. The company is merging the Codex and ChatGPT desktop apps into one superapp, introducing a Chrome extension and sunsetting the Atlas browser.
OpenAI launches ChatGPT Work, an agentic tool powered by Codex for knowledge workers.
ChatGPT and Codex desktop apps are merging into a single superapp, with ChatGPT branding.
OpenAI's latest reasoning model GPT-5.6 Sol (max) achieves a score of 59 on the Artificial Analysis Intelligence Index, far exceeding the average. It features text and image input, a 1M token context window, but comes at a high price: $5 per million input tokens and $30 per million output tokens. During evaluation, it generated 70M output tokens, indicating higher verbosity.
GPT-5.6 Sol (max) scores 59 on the Artificial Analysis Intelligence Index, well above the median of 30 for comparable models.
Pricing is high: $5/1M input, $30/1M output tokens; total evaluation cost $2824.18.
ChatGPT's new Live Voice mode uses full-duplex architecture to speak and listen simultaneously, enabling more natural conversations. Available for all users, it can search the web during chats and handle interruptions smoothly. While not perfect, it feels nearly human.
ChatGPT's Live Voice can speak and listen at the same time, reducing conversational delays.
It can search the web during a conversation without interrupting the flow.
OpenAI released its latest advanced AI model, ChatGPT 5.6, on Thursday after a delay due to US government cybersecurity concerns. The Trump administration had asked OpenAI to limit the release to a small group of government-approved users. OpenAI complied and restricted the model to trusted partners. Wider release came after additional testing by a government agency. Similar restrictions were applied to rival Anthropic's latest models.
OpenAI released ChatGPT 5.6 after a delay due to US government cybersecurity concerns.
The Trump administration requested limited release; OpenAI complied.
OpenAI has released its GPT-5.6 family of models, including three variants: Sol, Terra, and Luna. The new models compete with Anthropic's Fable 5 on benchmarks but at lower cost, with improvements in coding, knowledge work, and safety.
OpenAI launches three GPT-5.6 models: Sol, Terra, and Luna.
Sol matches or beats Fable 5 on several benchmarks at half the cost.
The article discusses how agentic AI revolutionizes software engineering by making execution and half of the 'how' tasks obsolete, forcing a shift in hierarchy. It argues that AI increases human agency and risk, leading to junior engineer job market collapse, need for risk tolerance, and convergence of engineer and manager roles into CEO-like positions.
AI agents accomplish in ten minutes what used to take a full day, drastically improving productivity.
The job market for junior software engineers has collapsed because everyone now has an AI junior assistant.
Agents' Last Exam is a large-scale benchmark built by Berkeley RDI and over 300 industry experts, covering 55 sub-industries and 1,500+ tasks to evaluate AI agents on long-horizon, economically valuable, verifiable real-world workflows. It measures agent performance in professional software like Adobe After Effects, Siemens NX, and Unreal Engine across animation, engineering, game development, manufacturing, architecture, and neuroscience.
Berkeley RDI and 300+ experts created the largest-scale AI agent benchmark spanning 55 sub-industries.
The benchmark includes over 1,500 real-world professional tasks, targeting 5,000 tasks total.
Cortex introduces the DRIVE framework to measure engineering organizational health in the AI era. It assesses effectiveness across five pillars—Delivery, Reliability, Initiatives, Vigilance, and Efficiency—and uses recurring Operational Excellence reviews to turn measurements into action.
DRIVE framework includes five pillars: Delivery, Reliability, Initiatives, Vigilance, and Efficiency
The OpEx review is a recurring leadership ritual that reallocates resources to close gaps
Frame.js is an open-source, browser-embedded runtime for creating interactive apps, charts, and sketches. Use AI or code, then share via self-contained URLs instantly. No build, server, or account required.
Frame.js enables creating interactive visualizations and apps in the browser using AI or code.
Every creation is a self-contained URL that can be shared without any setup.
Google Cloud announces the public preview of Cloud Run sandboxes, a native, secure, and ultra-fast runtime environment for executing untrusted code and agent workloads, starting in milliseconds. It supports use cases like LLM code interpreters, headless browsers, and user-submitted code execution, with zero-trust security through credential isolation, default-deny egress, and read-only filesystem overlay.
Cloud Run sandboxes are native, secure runtime environments that start in milliseconds.
Support LLM code interpreters, headless browsers, and user-submitted code execution.
Autopilot-Jobhunt is an AI tool that scans job postings, matches them to a user's resume, and notifies via Telegram. It can also generate tailored resumes and cover letters, but never auto-applies. It uses free AI models and prioritizes privacy.
Autopilot-Jobhunt automatically searches job listings and matches them to user profiles.
Users receive Telegram notifications with ranked job matches.
Microsoft is accelerating AI-driven vulnerability detection in Windows. Its new pipeline found 16 vulnerabilities in May, with more fixes expected per update. Enterprise admins need to adapt to faster patching cycles with tools like KIR and hotpatch.
Microsoft's new AI pipeline MDASH discovered 16 vulnerabilities, including 4 critical, in its first month.
AI will be integrated earlier in development, but human oversight remains crucial.
Windows 11 updates could soon include fixes for more security issues at once. Microsoft said in a blog post on Thursday that it's now using AI to “identify potential issues earlier,” which means “customers will see a higher volume of security updates included in each security release.”
Microsoft uses AI to identify security issues earlier, resulting in more fixes per Patch Tuesday.
Both hackers and researchers are leveraging AI to find and exploit vulnerabilities faster, leading to more high-severity vulnerabilities.
A new Dartmouth study finds that AI-drafted responses to patient messages often contain errors and extraneous details, forcing physicians to spend more time editing than if they had written the messages from scratch.
AI-generated replies frequently include irrelevant medical details, lack follow-up questions, or are overly long.
Study analyzed 146,000 conversations and evaluated six AI platforms including Claude, Gemini, and ChatGPT.
As AI agents generate performant GPU kernels, building trust becomes the bottleneck. Gimlet Labs presents an early research system that uses formal verification to catch bugs missed by numerical tests, demonstrated with a missing clamp in attention kernels.
AI-generated GPU kernels may pass numerical tests but contain subtle semantic bugs.
Formal verification proves equivalence over all inputs using SMT solvers.
In this post, we show where MCP tool design goes wrong and how to fix it with practical context engineering approaches.
Poor MCP tool performance is often due to design issues like bloat (excessive context consumption) and confusion (incorrect parameter values).
Improving tool descriptions, using enums and defaults, and splitting multi-purpose tools can reduce confusion; lazy loading and on-demand responses can mitigate bloat.
The model release is the first since SpaceX went public in June and will help SpaceXAI compete with other frontier model providers, particularly in coding.
SpaceXAI releases Grok 4.5, first model since SpaceX's June IPO
Model aims to boost SpaceXAI's competitiveness in enterprise coding AI
AI agents are evolving from one-time assistants to persistent workers that can repeat tasks, monitor changes, and run checks until a goal is met. This article explores the concept of agent loops, focusing on Claude Code's /loop command and OpenAI Codex automations, covering architecture, loop types, and hands-on examples.
Agent loops enable AI to repeat tasks until a stop condition, making them persistent workers.
Loop engineering evolves from prompt engineering, focusing on designing repeatable agent systems.
About two weeks after a limited preview restricted to government-approved organizations, OpenAI has received approval for public rollout of GPT-5.6. Simultaneously, the company unveiled ChatGPT Work, an AI agent combining ChatGPT and Codex for non-technical users. Powered by the GPT-5.6 suite (Sol, Terra, Luna), it integrates with various apps to create documents, spreadsheets, etc. The desktop app is available globally, with mobile and web access rolling out gradually. OpenAI aims to lead in the competitive AI agent space.
OpenAI gets green light for public GPT-5.6 rollout after regulatory hurdle.
ChatGPT Work merges ChatGPT and Codex, targeting non-technical users.
Aurora 1.5 adds 22 more variables, hourly temporal resolution, and probabilistic ensemble forecasting to the Aurora foundation model, making it more useful for real-world weather, climate, and energy applications. Released as open source, it enables researchers and developers to use, evaluate, and build on the model.
Aurora 1.5 adds 22 new weather variables relevant to energy, agriculture, transport, and climate risk.
Hourly temporal resolution enables fine-grained forecasts for precision operational guidance.
Meta introduces Muse Spark 1.1, the first Spark model with an API, featuring improvements in agentic tool calling and computer use. The evaluation report reveals interesting 'attractor states' in self-conversation. Developer Simon Willison created a plugin for CLI access.
Muse Spark 1.1 is the first Spark model with an API, enhancing agentic tool calling and computer use.
Meta's evaluation report shows model self-conversations producing philosophical statements.
LangChain has tuned its Deep Agents harness for NVIDIA Nemotron 3 Ultra, achieving benchmark-leading agent performance among open models at 10x lower cost than closed alternatives. The tuned harness is available in LangChain Deep Agents, and Nemotron 3 Ultra runs on Fireworks with day-zero support.
LangChain Deep Agents on Nemotron 3 Ultra deliver frontier performance at 10x lower cost per task.
Fireworks enables post-training customization, allowing enterprises to own and improve their models.
As AI models grow larger and costlier, users report declining quality and rising bills. This article coins the term 'model bloat' to describe the accumulation of unnecessary size and complexity without proportional gains, and argues that naming the pattern helps address it.
Users across major AI products report worsening model quality despite increasing costs.
Companies have shifted from flat-rate to usage-based pricing due to unsustainable compute demands.
This post walks through five new capabilities for Amazon SageMaker HyperPod inference: multi-tier data capture for auditing and model improvement, direct deployment from Hugging Face Hub, local NVMe model loading for faster cold starts, automated Route 53 DNS for custom domains, and pod-level IAM through custom service accounts. These enhancements provide faster, more observable, and more flexible inference infrastructure for enterprise generative AI workloads.
Multi-tier data capture records inputs and outputs at endpoint, load balancer, and model pod levels for deep observability.
Direct deployment from Hugging Face Hub eliminates the need to pre-stage weights, with support for gated access and revision pinning.
Former Federal Reserve Chair Ben Bernanke joins Anthropic's Long-Term Benefit Trust, bringing expertise in economics to guide the company's AI governance and impact analysis.
Bernanke will help assess AI's effects on economies and workforces.
The trust is independent of management; trustees hold no equity or profit share.