GitHub's September 22 changelog put Claude Opus 5.5 in the Copilot model picker for Pro+, Max, Business, and Enterprise across VS Code, Copilot CLI, the coding agent, JetBrains, and Xcode. In early testing it matched Opus 5 on task resolution while using "significantly fewer steps and tokens" and recovered from errors faster. It bills at provider list pricing under usage-based billing, so fewer tokens hits your invoice rather than a flat seat fee. Admins: new models auto-enable unless someone turned off the global default.

Copilot for JetBrains 1.18.0 shipped the same day with assisted approvals in public preview, organization and enterprise skills in both local and agent sessions, and message re-editing that rewinds the conversation plus the file changes. Try the re-edit rewind first. It turns a wrong agent turn into an undo instead of a git reset. Org-wide skills matter more at scale, because your conventions stop living in 30 engineers' local config. Check your Copilot plan →

antigravity-preview-09-2026 is live in the Interactions API and AI Studio, running on Gemini 3.8 Flash at unchanged pricing. Google's evals put it at 40% fewer output tokens on file edits and 30% lower cost on multi-turn coding, driven by better cache hits. Old requests still work, but tool names in step events changed (file tools now report as view_file, write_to_file, replace_file_content), so anything parsing step events needs a pass. Google deprecates antigravity-preview-05-2026 on October 5 and redirects its traffic to the new harness, so you get migrated that day whether your parser is ready or not.

A Credentials API landed alongside: store a secret once, attach it to an allowlisted domain or MCP server, and an egress proxy injects it at request time. Secret values are write-only and never returned, so a compromised agent cannot read back the tokens it is using. Eleven days, one string, a double-digit cost cut. Credentials ends the habit of pasting a bearer token into a prompt and trusting the sandbox. Antigravity agent docs →

Meta opened pre-orders at Connect on September 23 for its first audio-only AI glasses: no camera, 43 grams, up to 12 hours of music, calls, and assistant per charge plus 48 more from the case, prescription compatible, from $349, per Meta. Gen 3 is on sale now from $449 with a 12 MP camera, 3K video, and up to nine hours of battery. Meta VR Glasses land spring 2027 at $1,299.99, per Meta's Connect recap. Muse, Meta's agent, reaches the glasses "in the coming months."

A camera is the literal policy blocker on smart glasses in plenty of offices and clinics. For $349 that blocker is gone, 19 days out. Buy for the audio and the assistant, not for Muse, which is on no pair yet. Meta's announcement →

Claude Code for the solo dev who wants the top score on a hard change. OpenCode if you want the vendor-lock question to disappear, since it is MIT licensed and runs on more than 75 providers. Cursor only if the full IDE is why you showed up and month to month through November is acceptable. Current pricing plus the September LogRocket rankings put Claude Code at #1 and OpenCode at #2. Read the three-persona verdict → or try Cursor

Apertus 1.5, Switzerland's open sovereign model, became selectable inside Lumo on September 17, and in-chat charting keeps a compliance spreadsheet inside the encrypted session instead of exporting it. For a GDPR or HIPAA team of three to fifteen, Lumo for Business at $11.99 per user per month annually, per Proton, undercuts ChatGPT Business at $20 per Finout. Solo with no compliance requirement: stay on ChatGPT Plus or Claude Pro, because the privacy premium still does not pay for itself. Read the updated verdict → or try Proton Unlimited