GitHub Copilot Adds Claude Opus 5.5, GPT-6 Sol & Luna, and Grok 4.7 With Local Sandboxing
GitHub Copilot's weekly release notes are the big story in AI coding tools today: four new frontier models — Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 — are now available across paid Copilot plans, and a new local sandboxing feature limits what coding agents can touch on a developer's machine. Meanwhile, OpenAI's Codex CLI is heads-down in an active alpha cycle toward v0.159, and Claude Code and OpenCode both stayed quiet after yesterday's releases. Here's what changed.
GitHub Copilot Adds Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7
GitHub's September 21–25 weekly release notes confirm that Copilot users get four new frontier models to choose from. Claude Opus 5.5 and GPT-6 Sol are rolling out to Copilot Pro+, Business, Max, and Enterprise plans, while GPT-6 Luna and Grok 4.7 are available more broadly across Pro, Pro+, Max, Business, and Enterprise. That puts Anthropic's newest reasoning model, OpenAI's latest GPT-6 tier, and xAI's Grok 4.7 side by side in the same model picker for the first time.
Local Sandboxing Arrives in Public Preview
The standalone Copilot app now offers local sandboxing in public preview, letting administrators limit what an autonomous coding agent can reach on a developer's machine — restricting file system access, outbound network calls, and stored credentials. GitHub paired the feature with OpenTelemetry tracking so enterprises can pipe agent activity into their existing observability stack, a meaningful step for teams that have been asking for tighter guardrails before turning agentic autofix or cloud agents loose on production repos.
Also This Week: Slack, Teams, JetBrains, and VS Code 1.139
The rest of the week's Copilot changes spread across every surface developers touch. In Slack and Microsoft Teams, users can now switch models mid-conversation without losing per-thread preferences, and Copilot checks for duplicate issues before creating new ones. JetBrains IDEs get assisted approvals in public preview, which auto-authorizes low-risk agent operations, plus message editing with automatic session rewinding and organization-level skills that now carry over between local and agent sessions. VS Code 1.139 lets agents run inside Dev Containers over SSH, Tunnel, and WSL, alongside a more compact, filterable session list.
Codex CLI Grinds Through an Active v0.159 Alpha Cycle
OpenAI's Codex CLI shipped no new stable release in the last day, but the alpha channel is unusually active: the project cut alpha builds from 0.159.0-alpha.4 through at least alpha.8 in roughly 24 hours, following the stable rust-v0.157.x line that introduced GPT-6 Sol and Luna support with Amazon Bedrock routing. Judging by the pace of alpha releases, the team is working through Windows sandbox and app-server reliability fixes ahead of a v0.159 stable cut — worth watching if you run Codex on Windows.
Claude Code and OpenCode Hold Steady
Claude Code and OpenCode both stayed put. Claude Code remains on v2.1.283, the September 25 release that added model-locking controls (availableModelsMatch, deniedModels) and the /doctor prompt-audit command covered in yesterday's roundup. OpenCode is stable at v1.18.32, its fix for Bedrock image routing and Together AI usage reporting still the latest word from the open-source side.
What It Means for Developers
For most developers the actionable news today is on the Copilot side: if your plan includes Pro+ or higher, Opus 5.5 and GPT-6 Sol are worth a try for anything that needs longer reasoning chains, and Grok 4.7 is now an option even on the base Pro tier. If you administer Copilot for a team, local sandboxing in public preview is the guardrail to turn on before expanding agentic autofix or cloud agents to more repositories. Codex and Claude Code users have nothing new to install today, but Codex's alpha pace suggests a v0.159 stable release with Windows fixes is close.
Resources & References
Comments
Share your thoughts and join the conversation