Kimi K3 dominates Frontier Code v1.1 bench while performing poorly on Terminal Bench v4; differences between the two benchmarks explained

aa has this new Frontier Code v1.1 bench which measures Agentic Software Engineering. I see that Kimi K3 does very well in this new bench while performing badly in Terminal Bench v4. What are the differences between...