GitHub Copilot adds memory and new model choices for devs
Updated on December 29, 2025 4 minutes read
GitHub shipped a set of GitHub Copilot updates in December 2025 that focus on two themes: picking the right AI model for the job and reducing repeated context sharing inside a repository. The releases include a model picker for Copilot coding agent, general availability of auto model selection in Visual Studio Code, and a public preview of Copilot memory. For developers, the changes make performance and cost trade-offs more explicit, and help teams keep Copilot output closer to repository conventions.
What happened
On 8 December 2025, GitHub announced a model picker for Copilot coding agent for Copilot Pro and Copilot Pro+ subscribers. When starting an agent task, the dropdown includes Auto, Anthropic Claude Opus 4.5, Anthropic Claude Sonnet 4.5, and OpenAI GPT-5.1-Codex-Max.
On 10 December 2025, GitHub made Copilot auto model selection generally available in Visual Studio Code for all Copilot plans. In Auto, Copilot chooses a model based on real-time availability and can route to models such as GPT-5.1-Codex-Max, GPT-5 mini, GPT-4.1, Claude Sonnet 4.5, and Claude Haiku 4.5, depending on plan and policy.
GitHub also clarified how Auto affects premium requests. The 10 December 2025 changelog says Auto is limited to models with 0x to 1x premium request multipliers, and paid subscribers receive a 10% discount on the multiplier when using Auto (a 1x model draws 0.9 premium requests in Auto).
On 17 December 2025, GitHub said GPT-5.2 became generally available in GitHub Copilot via the chat model picker across supported clients. For Copilot Business and Copilot Enterprise, administrators must enable a GPT-5.2 policy in Copilot settings before users see it.
On 19 December 2025, GitHub introduced Copilot memory in public preview for Copilot Pro and Copilot Pro+. GitHub describes memory as repository-specific context that captures key insights over time and uses that shared context to improve Copilot coding agent and Copilot code review. Copilot memory can be enabled in GitHub Settings under Copilot.
Why it matters
Model choice is no longer a niche setting. The practical differences between models show up in latency, how well the assistant follows constraints, and how reliably it handles larger changes.
Auto model selection is aimed at reducing friction in the IDE. Instead of guessing which model is available, developers can let Copilot route requests based on capacity, then override manually when the task demands it.
Copilot memory targets consistency. Repeated prompts often end up restating the same rules about structure, tests, and tooling, and reviewers spend time correcting avoidable mismatches. A repository-scoped memory layer is designed to reduce that repetition and help agent and review workflows stay aligned with the codebase.
Key numbers
- Copilot Pro: 10 USD per month (or 100 USD per year), 300 premium requests per month.
- Copilot Pro+: 39 USD per month (or 390 USD per year), 1,500 premium requests per month.
- Copilot Free: 50 premium requests per month.
- Additional premium requests: 0.04 USD per request.
- Auto model selection in VS Code (GA on 10 December 2025): paid users receive a 10% multiplier discount, and Auto is limited to models with 0x to 1x multipliers.
- Selected multipliers for paid plans (GitHub docs): GPT-4.1 is 0x, GPT-5 is 1x, Claude Haiku 4.5 is 0.33x, Claude Opus 4.5 is 3x.
Context
GitHub's supported-models documentation includes a model retirement history with retirement dates and suggested replacements. That pace is one reason Auto matters: the list of models Copilot can route to changes over time.
Premium request billing began on 18 June 2025 for paid Copilot plans on GitHub.com, and counters reset on the first of each month at 00:00:00 UTC. GitHub's documentation also notes that Copilot coding agent consumes one premium request per session, multiplied by the model's rate, with additional consumption for real-time steering comments during an active session.
What's next
GitHub said Auto model selection is expected to evolve from availability-based routing to choosing the most appropriate model for a task based on complexity. GitHub also said the Copilot coding agent model picker is coming to Copilot Business and Copilot Enterprise, alongside additional integrations such as Visual Studio Code and GitHub Mobile.
For teams, a practical next step is to standardize a default. Many organizations will likely treat Auto as the baseline for Copilot Chat and reserve specific premium models for large refactors or review-heavy work.
How to go deeper
- Build day-to-day Git workflows and ship full-stack projects in the Web Development Bootcamp for US learners.
- Learn model evaluation basics and practical ML foundations in the Data Science & AI Bootcamp.
- Develop secure review habits and threat modeling skills in the Cybersecurity Bootcamp.
- Practice interview-ready explanations with Mock Interview Questions in the Learning Hub.