On May 17, 2026, GitHub made GPT-5.3-Codex the base model for Copilot Business and Copilot Enterprise, replacing GPT-4.1. The company also tied that switch to its first long-term support model, giving enterprise customers a stated availability window through February 4, 2027 just ahead of Copilot’s June 1 shift to usage-based billing.
What changed on May 17
The May 17 changelog turns GPT-5.3-Codex into the default model when an organization has not approved another model through its internal review process. GitHub says the change applies only to Copilot Business and Copilot Enterprise, not Copilot Pro, Pro+, or Free.
GitHub had already designated GPT-5.3-Codex as both the base model and the LTS model on March 18, but May 17 is the published enablement date. GitHub also says GPT-5.3-Codex launched on February 5, 2026 and will remain available through February 4, 2027 for Business and Enterprise customers.
That is a bigger deal than a normal model refresh. In GitHub’s enterprise setup, the base model is what Copilot uses when no other model is enabled, which means this is a policy and rollout change as much as a model change.
Why the LTS detail matters more than the model swap
Most coding-model coverage focuses on benchmarks. GitHub is making a different enterprise argument here: stability. An LTS window gives platform teams, security reviewers, and procurement owners a firmer planning horizon for extension testing, policy reviews, and internal rollout decisions.
That matters because Copilot is no longer just an autocomplete assistant. It now spans chat, CLI workflows, coding agents, cloud agent sessions, code review, and repository tasks. Once Copilot becomes part of a governed engineering workflow, sudden model churn becomes an operational risk, not just a developer annoyance.
GitHub’s documentation also shows how deliberate the rollout was. When a new base model is designated, organizations get a 60-day upgrade window before automatic enablement. That is a control-plane pattern: model management is becoming part of enterprise software administration, not just product marketing.
Operational impact for engineering leaders
There are three practical implications behind the headline.
- Default behavior changes at the admin layer. If an organization has not approved other models, Copilot now lands on GPT-5.3-Codex first.
- Client compatibility still matters. GitHub lists minimum version guidance for GPT-5.3-Codex across major tools, including Visual Studio Code v1.104.1 or later, Visual Studio 17.14.19 or later, JetBrains IDEs 1.5.61 or later, Copilot CLI 0.45.0 or later, and Xcode 0.13.0 or later.
- Cost governance is about to get sharper. GPT-5.3-Codex currently carries a 1x premium request multiplier, while GPT-4.1 remains force-enabled at a 0x multiplier for now.
That last point matters because GitHub is moving Copilot to usage-based billing on June 1, 2026. Premium request units will be replaced by GitHub AI Credits, and GitHub says fallback experiences available under the current request model will go away under the new credit system. For business customers, that means model choice is about to affect finance, policy, and adoption planning much more directly than before.
GitHub’s billing guidance keeps Copilot Business at $19 per user per month and Copilot Enterprise at $39 per user per month, with higher promotional included usage during the initial transition window. That gives enterprises a short runway to measure whether longer agent runs and heavier GPT-5.3-Codex usage become a meaningful budget line.
What to watch next
The next key date is June 1, 2026. That is when GitHub’s usage-based billing begins and when GPT-4.1 is set to deprecate alongside that shift. The combination matters more than either update on its own: GitHub is tightening the link between default model policy, cost visibility, and enterprise rollout discipline.
The broader signal is that coding-agent platforms are starting to look more like governed infrastructure than consumer AI products. Enterprises increasingly want a default model they can trust, a support window they can plan around, and billing controls that keep agent usage from becoming a surprise spend category.
For AI agents and automation more broadly, this is the same pattern showing up across the market: the winning platforms are not just shipping better models. They are shipping clearer control planes around model selection, rollout timing, budgets, and operational guardrails.