🎉 December 2025 Copilot Roundup

[Akash1134]

🎉 December 2025 Copilot Roundup

Welcome to 2️⃣ 0️⃣ 2️⃣ 6️⃣ December wrapped a huge Copilot year with a new flagship model, broader sharing, and billing changes you’ll want to lock in.

Think of 2025 as Copilot’s growth spurt: new GPT‑5.x/Opus 4.5 brains, faster defaults, and smoother replies. Agents graduated into mission control with custom agents and skills so you can launch work in parallel. Spaces took over from old knowledge bases, letting you share context and pull code straight from the viewer. Admins got BYOK, real spend controls, and data residency. And Copilot showed up everywhere—IDE, CLI, web, and mobile—with clearer model and plan pickers. Want all the top posts of 2025 in one place? See Agentic AI, MCP, and spec-driven development — top blog posts of 2025.

---

⚡ This Month in Copilot

• GPT-5.1-Codex-Max public preview — Now selectable in chat/ask/edit/agent, on github.com, mobile, and CLI. Enterprise/Business must enable the policy; Pro/Pro+ can self-opt in. GitHub changelog
• Public & individually shared Spaces — Share personal spaces publicly or with specific users; viewers only see files they’re authorized for. Add files straight from the code viewer. Changelog
• Spaces code-viewer entry point — Add files to Spaces directly from the github.com code viewer for faster grounding. Changelog
• VS model-picker updates — Refined model selection including auto-pick options and new models. Changelog
• Older models deprecated (Oct 23) — Some legacy OpenAI, Anthropic, and Google models were retired in Copilot, with suggested replacements listed in the official model retirement history. GitHub Docs

Previously in November

---

🏆 Copilot 2025: What We Achieved

• Models & accuracy: Two GPT‑4o upgrades (March & May) improved completions by default; year-end brought GPT‑5.1/5.1‑Codex‑Max previews plus GPT‑5.2 and Claude Opus 4.5, while legacy models were retired with suggested replacements tracked in GitHub Docs. March upgrade · May upgrade · Model retirement history · GPT‑5.1‑Codex‑Max · Opus 4.5 GA
• Agents everywhere: Mission control launched so you can assign, steer, and track Copilot coding agent tasks; Agent Skills arrived; custom agents expanded across JetBrains, Eclipse, and Xcode; Copilot CLI gained semantic search. Mission control (changelog) · Mission control (guide) · Agent skills · Custom agents in IDEs · CLI semantic search
• Spaces replace knowledge bases: Spaces introduced in May, GA in September, migration tool in October, public/individual sharing and code-view add in December; knowledge bases sunset on November 1. Launch · GA · Migration · Sunset notice · Public sharing
• Enterprise-grade control: BYOK preview let enterprises bring Anthropic/OpenAI/xAI/Foundry keys; premium-request budgets/policies replaced $0 budget workarounds; data residency added for code review previews. BYOK · Premium policy · Data residency
• Editor experience: Auto model selection expanded in IDEs; next-edit suggestions matured in JetBrains; repository custom instructions kept teams consistent across sessions and PRs. VS model picker · Next-edit suggestions · Repository instructions (Docs) · Copilot code review playbook (repo)
• Surface & plan expansion: Copilot plans and model availability expanded across GitHub.com, IDEs, mobile, and CLI, with official docs tracking what’s included per plan and what’s available per client. Plans for GitHub Copilot · Supported models

---

📚 Noteworthy Reads

• Debug issues faster with Copilot Spaces — A step-by-step guide to collecting repo + issue context and then triggering Copilot coding agent from the space. GitHub Blog
• WRAP up your backlog with Copilot coding agent — A framework for writing good issues, refining instructions, and breaking work into atomic tasks for higher-quality PRs. GitHub Blog
• Premium requests and budgeting — How premium requests work (including dedicated SKUs for Spark and Copilot coding agent starting Nov 1, 2025) and what happens if you run out. GitHub Docs
• Upcoming removal of $0 premium request budgets (Enterprise/Team) — Background on the Dec 2, 2025 change referenced in the docs. GitHub Changelog
• Mission control: orchestration guide — A practical walkthrough of running multiple Copilot coding agent tasks in parallel, steering mid-run, and reviewing session logs. GitHub Blog
• Custom agents ecosystem — Introduces partner-built custom agents and how to define your own at .github/agents/. GitHub Blog

Thanks for building with us through 2025. We’ll keep shipping steady improvements - check the Copilot label in the GitHub Changelog when you want the latest. Onward through 2026. 🎆

[MasteraSnackin]

Great roundup! 2025 was definitely a huge year for Copilot. I'm particularly excited about the Mission Control feature - being able to orchestrate multiple agent tasks in parallel is a game changer for handling larger refactoring work.

The GPT-5.1-Codex-Max preview looks promising too. Has anyone tested it extensively yet? Curious to hear how it compares to the previous models in terms of code quality and understanding context.

Also really appreciate the move from knowledge bases to Spaces with the public/individual sharing options. Makes it much easier to collaborate with specific teammates without giving access to everything.

Looking forward to seeing what 2026 brings! 🚀

[lkraav]

In early 2026, I'm wondering about the subpar UX of Copilot web coding agent.

It takes something like 90+ sec minimum for it to spin up its environment, and if it shuts down before you manage to complete typing to send steering instructions, you now have to wait through another cold boot. Repeat 10-20 times in a session, yeah no thanks.

This stop-n-go nature, with such stark contrast to always-on local IDE-based agent use, pushes this current web-app solution iteration to untolerable territory.

I like that in GH webapp Copilot session is available to the whole team, so anyone can go in, investigate details or learn things, and move things along towards PR completion in a shared fashion.

But in reality, in early 2026, it seems only to be good enough to

• human files initial issue, assigns to
• comes up with an initial solution PR
• someone finishes it up locally, avoiding stop-n-go work pattern, but also somewhat losing some of solution transparency compare to if it all would've been worked through in the open session

Am I missing something important on how to use Copilot Coding Agent in GH webapp?

Tried search for existing discussions with something like https://github.com/orgs/community/discussions/categories/copilot-conversations?discussions_q=category%3A%22Copilot+Conversations%22+spinning+sort%3Adate_created but AFAICT, this hasn't been discussed here. I can open a full new topic if so directed.

[lkraav]

OK re-posting on its own for better visibility.