Copilot Cloud Agent adds cheaper “economy” models — Claude Haiku 4.5 and GPT‑5.4 mini

GitHub added two cheaper model options to Copilot cloud agent on May 18: Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.4 mini. The company said both carry a 0.33x multiplier, positioning them as “fast, cost-efficient models for simple tasks” inside the coding agent workflow. Copilot cloud agent runs in an ephemeral development environment powered by GitHub Actions, where it can inspect code, make changes and run tests before opening a pull request for review. That product change lands just before GitHub’s June 1, 2026 move from request-based billing to usage-based billing for Copilot. GitHub’s documentation says chat, CLI, cloud agent, Spaces, Spark and third-party coding agents will consume AI credits, with charges tied to model choice and token usage. A quick interaction with a lightweight model can cost a fraction of a credit, GitHub says, while a longer coding-agent session on a frontier model costs more. (github.blog) ### Why did GitHub add cheaper models to cloud agent now? GitHub’s own changelog said the new options are meant for “simple tasks,” while larger models remain available for more complex work. The company framed the update as a way to “pick the right model for the job,” with smaller models for straightforward changes and more capable models for harder tasks. (docs.github.com) GitHub’s model-selection documentation also shows that cloud agent already supports switching models when users assign an issue to Copilot, mention `@copilot` in a pull request comment, or start a task from supported agent interfaces. Where no picker is available, GitHub says “Auto” is used. (github.blog) ### What do Claude Haiku 4.5 and GPT-5.4 mini represent inside Copilot? Anthropic introduced Claude Haiku 4.5 in October 2025 as a smaller model aimed at faster, cheaper performance, saying it offered coding performance similar to an earlier frontier model at one-third the cost and more than twice the speed. GitHub rolled Haiku 4.5 into Copilot in late 2025 and later made it generally available across supported IDEs. (docs.github.com) OpenAI’s GPT-5.4 reached general availability in GitHub Copilot on March 5, 2026, and GitHub made GPT-5.4 mini generally available for Copilot on March 17. GitHub described GPT-5.4 as its latest “agentic coding model,” while GPT-5.4 mini became another selectable lower-footprint option in the Copilot model picker. (anthropic.com) ### Why are teams talking more about cost and control? GitHub’s pricing pages make model choice more consequential because organizations will be billed by usage and model rate rather than by a flat request count. The documentation says per-token pricing and model multipliers determine how many AI credits are consumed across paid Copilot features. (github.blog) An XDA Developers article published this week described one developer replacing GitHub Copilot with a self-hosted setup, citing tighter limits, changing economics and a preference for local control. That account is anecdotal, but it reflects a broader pressure point as coding assistants move toward metered usage and teams weigh predictability, governance and data location alongside model quality. (docs.github.com) ### Where does this leave enterprise users? GitHub’s documentation says enterprise and business administrators can enable or restrict specific models through Copilot settings, and some organizations can limit users to FedRAMP Moderate-certified models in U.S. data-residency environments. Those controls make model selection a governance decision as well as a cost decision. (xda-developers.com) Starting June 1, 2026, those choices will matter more. GitHub’s billing documentation says usage-based charging will apply across Copilot’s paid agent and chat surfaces, and the cloud-agent model menu now includes cheaper options for teams deciding when a smaller model is enough. (docs.github.com) (github.blog)