it seems like 5.6 SOL is better at almost everything than Mythos except Coding Benchmarks (except TerminalBench)? anyone knows why Mythos scores so high on SWEBench are they cheating or are they just optimised better for coding?