GitHub Launches Copilot Fusion, Closely Matches Claude Code in Benchmark

ChainCatcher report, according to Beating, GitHub officially launched the multi-model scheduling system Copilot Fusion at the Universe 2026 conference. Simple tasks are assigned to appropriate models; complex tasks are first attempted by lower-cost models, and if the results are insufficient, the system switches to a more powerful model or enables cross-checking between models. Independent evaluation agency Artificial Analysis found that Fusion Max achieved a programming composite score of 60, close to Claude Code paired with Opus 5.5 Max’s score of 62. Fusion Max averaged $5.59 in API cost per task and took approximately 36 minutes; Claude Code with Opus 5.5 Max cost $11.42 and took about 60 minutes, representing a roughly 51% reduction in cost and a 40% reduction in time. The evaluation showed Fusion’s cache hit rate at 97%, compared to 95% for Claude Code. The original article also noted that frequent model switching may cause cache losses; these results reflect only specific benchmark tests and do not imply that all tasks in real-world usage will achieve a 50% cost reduction.