AI Coding News: July 29, 2026 — Codex CLI Goes Stable, Grok 4.5 Joins Copilot
OpenAI's Codex CLI finally tagged a stable release after eleven alpha builds in five days, OpenCode shipped two more patch releases to shore up Model Context Protocol compatibility, and xAI's Grok 4.5 landed inside GitHub Copilot as a model option that undercuts Claude Opus 5 on price. Here is everything that shipped across the major AI coding tools on July 29, 2026.
OpenAI Codex CLI Reaches Stable v0.146.0
After pushing eleven unstable alpha builds in the five days leading up to July 26, OpenAI's Codex CLI tagged a stable v0.146.0 release (install with npm install -g @openai/[email protected] ). The release adds session naming via /new or /clear, manifest support for Agent Plugins, and the ability to fork threads with paginated history — including temporary forks that never appear in the thread listing. On the reliability side, it fixes proxy configuration across authentication and plugin downloads, cleans up MCP connection management, and improves terminal responsiveness in long-running conversations.
This builds directly on v0.145.0 from July 21, which introduced experimental paginated thread history, a /import command for migrating settings from Cursor and Claude Code, Amazon Bedrock login support, and GPT-5.6 Sol as the default model. The gap between that release and today's stable tag is the story: OpenAI spent over a week hardening the same feature set in alpha before committing to it.
OpenCode v1.18.9 Patches MCP Legacy Compatibility
The OpenCode team shipped two releases on July 28. v1.18.8 enhances MCP server compatibility, reconnects servers after expired SDK sessions, and honors OAuth callback port configuration. v1.18.9 followed the same day, restoring compatibility with legacy MCP SDK clients and fixing a Solid cleanup crash that affected desktop navigation, alongside a new collapsible model-provider sections UI and a V2 desktop sidecar option.
The timing lines up with the Model Context Protocol's final 2026-07-28 specification revision — the protocol's largest rewrite since launch, covering a stateless core, a formal Extensions framework, and hardened OAuth. OpenCode's rapid-fire patches are effectively day-one compliance work against that new spec.
Grok 4.5 Debuts in GitHub Copilot
xAI's Grok 4.5 is now rolling out inside GitHub Copilot, built for fast, agentic coding and complex multi-step workflows. It ships with a 500,000-token context window, text and image input support, and low/medium/high reasoning-effort settings, and performs particularly well when dispatching tools in parallel. It's available on Copilot Pro, Pro+, Max, Business, and Enterprise plans across VS Code, Visual Studio, Copilot CLI, the Copilot cloud agent, the Copilot app, JetBrains, Xcode, and Eclipse — though Enterprise and Business admins must explicitly enable the Grok 4.5 policy, since it's disabled by default. It's billed at provider list pricing under usage-based billing, which multiple early reports peg at roughly a 76% discount versus Claude Opus 5 for comparable coding workloads.
The same day, GitHub also extended Copilot app usage metrics across report rollups for better enterprise visibility — a smaller change, but one that matters for organizations tracking AI spend across a growing list of model options.
What This Means for Developers
The multi-vendor coding stack is now the default, not the exception — recent industry surveys put the share of developers running three or more AI coding tools in parallel at 59%. A typical setup pairs Cursor for editing, Claude Code for architecture and debugging, and Copilot for in-editor completions. Grok 4.5's price-to-context ratio makes it a strong candidate for high-volume, exploratory tasks where you don't want Opus-level billing; Codex's stable tag ends weeks of alpha churn and gives teams a version worth pinning in CI; and OpenCode's compatibility fixes are the kind of unglamorous but necessary work that keeps MCP-based agent tooling from breaking as the protocol itself evolves underneath it.
Resources & References
Comments
Share your thoughts and join the conversation