AI & app changelog

From new models and portable plugins to smoother agent workflows, this week’s updates make GitHub Copilot more flexible across editors, the command line, and the Copilot. Source: GitHub Copilot.

Introducing Gemini 3.7 Flash. Source: DeepMind.

OpenAI opened a limited preview of an “Ultrafast” service tier that runs GPT-5.6 Sol on Cerebras inference hardware at up to roughly 14x the usual speed — around 750 output tokens per second. It targets latency-sensitive agent loops where waiting on the model is the bottleneck. As a preview it has no published price, no general-availability date, and it is an API service tier rather than a switch inside the Codex UI yet.

The ChatGPT desktop app — where Codex now lives — gained a Linux preview covering Ubuntu, Debian and Fedora, closing the last major desktop gap after macOS and Windows. Alongside it, the Codex CLI added an /import command that pulls your existing setup, settings and recent work in from Claude Code, Claude Cowork and Cursor, lowering the switching cost from rival agents.

OpenAI launched GPT-5.6-Cyber, a model specialised for cybersecurity work, and gated it behind a new vetted access tier called “Daybreak Red” for defenders who clear its screening. Daybreak-enrolled Codex accounts also switch to a safer default “auto-review” mode instead of full-access execution, reflecting the extra caution around a security-tuned model.

This week’s updates across GitHub Copilot in the desktop app, CLI, and VS Code help you resume and organize work, review changes, and ask questions without losing. Source: GitHub Copilot.

Codex CLI 0.147.0 introduced portable “Agent Plugins” — installable extensions discoverable across a local, workspace and remote catalog — plus an –approve-for-me flag that auto-approves actions for hands-off runs, and support for the 2026-07-28 MCP protocol (paginated discovery, multi-round requests). The deprecated –full-auto flag was removed; use –sandbox workspace-write instead.

OpenAI set August 31, 2026 as the retirement date for GPT-5.4 and GPT-5.4-mini for users who authenticate Codex with a ChatGPT plan. The official migration path is GPT-5.4 to GPT-5.6-Terra and GPT-5.4-mini to GPT-5.6-Luna. The older models remain reachable through API-key authentication, so scripted and API workflows are not forced off on the same date.

Explore lower GPT‑5.6 pricing for Luna and Terra—and how OpenAI’s more efficient models help enterprises deploy AI workflows at scale. Source: OpenAI.

This changelog covers VS Code v1.127 through v1.131, shipped throughout July 2026. These releases improve how you work with agents, review changes, use chat, and. Source: GitHub Copilot.

Codex CLI 0.146.0 added session naming and pinning, Agent Plugins manifest support, paginated thread history with temporary forks, and a remote “Code Mode” over WebSocket for driving sessions from elsewhere. Separately, a “Sign in with ChatGPT” beta began rolling out on partner platforms — Airtable, GitLab, HubSpot, Notion, Supabase and Vercel among them.

Grok 4.5, xAI’s latest reasoning model, is now rolling out in GitHub Copilot. It is designed for fast, agentic coding and complex multi-step workflows, with a context. Source: GitHub Copilot.

Health in ChatGPT now lets eligible U.S. users securely connect medical records and Apple Health to get more personalized insights and better understand their health. Source: OpenAI.

Introducing OpenAI Presence, a proven enterprise AI agent platform that helps organizations deploy trusted voice and chat agents for customer and internal workflows. Source: OpenAI.

We’re introducing new Gemini models, including Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber. Source: DeepMind.

You can now manage a cost center’s AI credit pool directly in the billing UI where you create and edit cost centers. Previously, you could only manage this through the…. Source: GitHub Copilot.

Pro and Team Standard move to usage credits (one-time $100 credit), then Fable 5 at API rates ($10/$50 per 1M tokens). Standard limits bonus phase also ends. Source: Claude.

Google introduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model to find and patch vulnerabilities. Source: DeepMind.

The GPT-5.6 family reached general availability across ChatGPT, Codex and the API: Sol (flagship for complex coding and computer use), Terra (balanced, roughly GPT-5.5-level at lower cost) and Luna (fastest and cheapest). All three carry a ~1.05M-token context window, and the naming replaces the older “…-Codex” model suffix — one family now powers every Codex surface.

A new generation of voice models for natural human-AI interaction, now powering ChatGPT Voice. Source: OpenAI.

Introducing computer use in Gemini 3.5 Flash. Source: DeepMind.

ChatGPT Pro $200/mo tier confirmed for unlimited reasoning.

Claude Opus 4.6 rolled out to Pro subscribers.

~150 msgs/3h cap observed on current usage.

5-hour rolling message window confirmed.

Gemini Advanced renamed to Google AI Pro as part of the wider One AI plan restructuring.

Annual plan discount structure changed.

New per-day file upload limits introduced.

Codex pushed past pure coding. OpenAI added six role-specific plugin packs (data analytics, sales, product design, creative, and two finance verticals — 62 apps and 110 skills), “Codex Sites” for hosting internal dashboards, and annotations that now extend to docs, spreadsheets and slides rather than just code. OpenAI reported 5M+ weekly Codex users, with non-developers growing about 3x faster than developers.

AI features bundled into the Business tier.

A full-featured Codex arrived inside the ChatGPT mobile app on iOS and Android — available on every plan, including a Free/Go preview. You can review generated code, approve or reject actions, switch models and redirect a task, all against a session hosted on your Mac. Windows-host control was “on the roadmap” at launch.

OpenAI shipped a Codex extension for Chrome — a browser-native agent that can test web apps, use DevTools, hold multi-tab context and operate sessions you are already logged into. OpenAI cited 4M weekly active users at the time, about 8x growth since the start of the year.

A major macOS update gave Codex “Computer Use” — it operates Mac apps and the browser with its own cursor and can run several agents in parallel without disturbing what you are doing — plus image generation, persistent memory of your preferences and stack, and 90+ new plugins (Atlassian Rovo, CodeRabbit, CircleCI, GitLab and more). Windows parity for these was promised “in the coming weeks”.

OpenAI replaced Codex’s flat per-message metering with API-token-aligned usage credits (about $0.04 per credit) consumed inside rolling 5-hour windows, across Plus, Pro, Business and Enterprise. A typical task runs 5-45 credits. It is the period’s biggest structural change — and the source of most “my quota evaporated in minutes” complaints, since heavy agent runs burn credits fast.