GitHub adds model selection

GitHub now lets people choose Claude and Codex models for coding agents on github.com instead of taking a single default. (github.blog) GitHub said the change went live on April 14, 2026, for the Claude and Codex third-party coding agents on github.com. The company said users can pick Anthropic models for Claude tasks and OpenAI models for Codex tasks when they start a job. (github.blog) The new picker mirrors a feature GitHub already offered for its own Copilot cloud agent. GitHub said the goal is to give users access to newer and more capable models as those models are added. (github.blog) A coding agent is software that can take a task, inspect a repository, write code, and open a pull request with the result. GitHub’s update shifts one more part of that workflow from the model maker to the platform that hosts the repository and assigns the work. (github.blog) GitHub’s own documentation already treats model choice as a product setting with tradeoffs in speed, accuracy, and cost. The company says some models have lower latency, some hallucinate less, and some consume more of a user’s monthly premium-request allowance through higher multipliers. (docs.github.com 1) (docs.github.com 2) That matters because GitHub has spent the past year turning Copilot from a single assistant into a menu of models, limits, and billing controls. GitHub says paid plans get a monthly premium-request allowance, counters reset on the first day of each month at 00:00 Coordinated Universal Time, and extra usage can be billed if overages are enabled. (docs.github.com 1) (docs.github.com 2) The company has also been tightening those controls as demand rises. On April 10, 2026, GitHub said it was enforcing new limits for Copilot Pro+ and retiring Opus 4.6 Fast, citing service reliability and a plan to focus resources on the models customers use most. (github.blog) The wider market is moving the same way: one interface, several model providers, and a picker that decides which engine handles each job. Microsoft’s Visual Studio Code said on February 5, 2026 that users could run Claude and Codex agents alongside GitHub Copilot in the editor, with local and cloud options. (code.visualstudio.com) OpenAI has been reshuffling the Codex side of that stack too. In its Codex changelog, OpenAI said that as of April 14 users signing in with ChatGPT could choose from gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, and gpt-5.2, while other API-supported models require an API key or another configured provider. (developers.openai.com) GitHub’s update does not turn model choice into an abstract research problem; it puts the choice next to the repository, the quota, and the bill. On github.com, the platform is increasingly deciding how developers reach Anthropic and OpenAI models in the first place. (github.blog) (docs.github.com)