According to CoinMeta, GitHub officially launched the multi-model scheduling system Copilot Fusion at the 2026 Universe Conference. This system is capable of assigning simple tasks to the appropriate models, while more complex tasks can first be attempted by less expensive models. If those attempts are unsuccessful, it will switch to more powerful models, or different models can be used to check each other's work. Tests conducted by the independent evaluation institution Artificial Analysis showed that Fusion Max scored 60 in comprehensive programming, which is close to the 62 points achieved by Claude Code in combination with Opus and Max. Fusion Max costs an average of $5.59 per task and takes about 36 minutes, whereas the combination of Claude Code and Opus costs $11.42 and takes about 60 minutes. The cost of Fusion was reduced by about 51%, and the time taken was shortened by about 40%. Although frequent model switching may result in cache losses, in this evaluation, the cache hit rate of Fusion reached 97%, and that of Claude Code was 95%. However, these results only represent specific benchmark tests and do not mean that half of the time can be saved for all tasks in actual use.