I Built KingBench 4 (HARD): Opus 5.5 vs GPT-6 Astra vs Sol vs Haiku 5.5 vs GLM 5.3

In this video, I'll put GPT 6.1 Sol and GPT 6 Astra through Codex, Opus 5.5 and Haiku 5.5 through Claude Code, and GLM 5.3 through OpenCode against the same six projects: a Game Boy, DS and 3DS emulator, a Docker alternative, a 3D room designer, a VIA alternative for QMK keyboards, a Pokémon Red recreation through the first gym, and a Fitbit desktop app with a Rust backend. I'll go through what each model built, where the apps broke, and how each one was scored out of ten under the same strict rubric. Then we'll look at the final chart across all six projects and at the tokens and API-equivale