GPT-5.6 Launches With Sol, Terra, and Luna
On July 9, OpenAI broadly launched the GPT-5.6 family, led by Sol and accompanied by the cheaper Terra and Luna variants. OpenAI pitched the release around stronger performance per dollar, more polished computer use, and better results across coding, knowledge work, cyber, and science, while also introducing "ultra" as its highest-capability setting for the hardest tasks.
ChatGPT Work Turns OpenAI's Coding Stack Into a White-Collar "Super App"
Also on July 9, Reuters reported that OpenAI launched ChatGPT Work, combining ChatGPT with Codex so non-coders can create documents, presentations, and websites from one workspace. The move put OpenAI more directly against Anthropic's Claude Cowork and Microsoft's Copilot Cowork, and showed the next battleground is not just model quality but who owns the everyday interface for AI work.
Alberta Turns Claude Into a Government Cybersecurity Workhorse
On July 6, Anthropic said Alberta's Ministry of Technology and Innovation had used Claude Code with Opus and Sonnet to scan 466 million lines of code in 20 hours, surface vulnerabilities across government systems, and build internal tools to harden them. The story mattered because it showed Claude moving beyond pilot projects into large-scale public-sector security work.
CISA Starts Using Anthropic's Mythos to Audit U.S. Government Code
Reuters reported on July 6 that CISA's Attack Surface Evaluation team is using Mythos to scan government repositories for bugs that could be exploited by cybercriminals or foreign intelligence services. The reported audits had already uncovered many vulnerabilities, making Mythos one of the clearest examples yet of a frontier cyber model being used inside real federal security workflows.
Britain's Financial Watchdog Signals AI Models May Need Direct Oversight
Also on July 6, a senior FCA official said Britain should consider regulating large language models like ChatGPT, Claude, and Gemini as general-purpose AI tools because they are increasingly shaping consumer financial decisions. The warning highlighted how quickly general models are moving from optional assistants into infrastructure that regulators may feel forced to supervise directly.
Google Makes Gemini Managed Agents More Practical for Real Production Work
On July 7, Google added background execution, remote MCP server integration, custom functions, and credential refresh to Managed Agents in the Gemini API. The upgrade pushed Gemini further from demo-friendly agents toward asynchronous workers that can run longer tasks inside connected environments without constantly blocking the application that called them.
GPT-Live Pushes OpenAI's Voice Stack Toward Real-Time Conversation
On July 8, OpenAI introduced GPT-Live, a new voice family built to listen and speak simultaneously and now rolling out globally in ChatGPT Voice. OpenAI framed it as a step toward more natural, longer-running, and more agentic spoken interaction, making voice feel less like turn-taking transcription and more like a live conversational interface.
