Anthropic's Claude Opus 4.1 achieves 74.5% on coding benchmarks, leading the AI market, but faces risk as nearly half its $3.1B API revenue depends on just two customers.