← back to jyje/pilot-upstage-solar-open2

Gallery

Real apps that Upstage's Solar models actually built through this repo's agent-harness cases — played and functionally checked in a real browser, not just a text reply. Nothing here is hand-written or hand-edited beyond a couple of openly-noted, minimal bugfixes (see each entry). Each entry names the model that built it: the same spec given to two different models does not produce the same result.

6×6 Mini Sudoku — solar-open2

Built by solar-open2 through omp (oh-my-pi), asked only for a written requirements spec — puzzle generation, live conflict highlighting, real win detection, and a working "New Puzzle" button, all self-contained in one HTML file.

Case 08 — omp solar-open2

6×6 Mini Sudoku — solar-pro4

The same spec, same harness, same 8-minute budget — given to solar-pro4 instead, as a head-to-head. Its raw output had one hard syntax error that killed the whole script; handing that exact line back fixed it in a single round, after which it passed all six browser checks.

Case 08 — omp solar-pro4