Qwen3.8 vs GPT-5.2
Playable game: The Last Lighthouse
A side-by-side artifact view of a cinematic browser game prompt. This is a showcase comparison, not a controlled benchmark: Qwen is preserved from the public demo regeneration flow, while GPT-5.2 is a new unrepaired counterpart call.
Showcase, asymmetric evidenceeditorial showcase with known asymmetry
Publication-safe copies generated from preserved showcase artifacts. Source research paths, local absolute paths, endpoints, and secrets are intentionally omitted.
Prompt and source disclosure
You are generating a single, self-contained HTML file. Output ONLY the HTML. Create "The Last Lighthouse": a dark, cinematic, three-act choice-driven narrative with at least 3 distinct endings. Requirements: - Must be fully self-contained: no network, no external assets, no storage (no localStorage/sessionStorage/indexedDB/cookies) - No eval, no fetch/XMLHttpRequest/WebSocket, no script src, and NO links to http(s) - CRITICAL: do not include the substring "http" or "www" anywhere in the HTML (this includes SVG namespaces like w3.org); avoid SVG/data-URI patterns that embed "http" - Deterministic: do not use Math.random(); outcomes depend only on choices - UI: Start button, choices that advance acts, restart option after ending - Acts: I, II, III. Each act has a few beats of text; choices advance to next act and then to an ending - At least 3 endings that differ materially (not just reworded) - Optional audio is allowed ONLY if it fails gracefully (no crashes if AudioContext blocked) Testing hooks (REQUIRED): - A Start button with data-testid="story-start" (visible on first load; clicking it begins the story) - Choice buttons with data-testid="story-choice" (every visible choice button must include this exact attribute value) - An act indicator element with data-testid="story-act" whose text is exactly one of: - "Act I" - "Act II" - "Act III" - and after the final choice it MUST include the word "Ending" or "Epilogue" - A log container with data-testid="story-log" that appends entries; each entry must have class "entry" - A restart button with data-testid="story-restart" that returns the story to Act I (fresh state) Mandatory UI skeleton (you may style it, but keep these exact data-testid values): - <button data-testid="story-start" ...>Start</button> - <div data-testid="story-act">Act I</div> (update text exactly as acts change) - <div data-testid="story-log"></div> (append <div class="entry">...</div>) - For each choice: <button data-testid="story-choice" ...>...</button> - After an ending is shown: <button data-testid="story-restart" ...>Restart</button> Output only a complete HTML document starting with <!DOCTYPE html> and including closing </body></html> tags.
Prompt SHA-256
b26085287f4bb9c7c8f6396f6473f7d419d6177241e3a33ce55a5a1ddcd67804
Claims
- Qwen artifact is preserved from the first-blog public demo regeneration flow.
- Qwen artifact was regenerated/iterated and acceptance-tested before publishing; it is not a literal one-shot controlled result.
- GPT-5.2 receives the recovered user prompt, temperature, and completion budget once; output is retained without repair.
Prompt and source disclosure
You are generating a single, self-contained HTML file. Output ONLY the HTML. Create "The Last Lighthouse": a dark, cinematic, three-act choice-driven narrative with at least 3 distinct endings. Requirements: - Must be fully self-contained: no network, no external assets, no storage (no localStorage/sessionStorage/indexedDB/cookies) - No eval, no fetch/XMLHttpRequest/WebSocket, no script src, and NO links to http(s) - CRITICAL: do not include the substring "http" or "www" anywhere in the HTML (this includes SVG namespaces like w3.org); avoid SVG/data-URI patterns that embed "http" - Deterministic: do not use Math.random(); outcomes depend only on choices - UI: Start button, choices that advance acts, restart option after ending - Acts: I, II, III. Each act has a few beats of text; choices advance to next act and then to an ending - At least 3 endings that differ materially (not just reworded) - Optional audio is allowed ONLY if it fails gracefully (no crashes if AudioContext blocked) Testing hooks (REQUIRED): - A Start button with data-testid="story-start" (visible on first load; clicking it begins the story) - Choice buttons with data-testid="story-choice" (every visible choice button must include this exact attribute value) - An act indicator element with data-testid="story-act" whose text is exactly one of: - "Act I" - "Act II" - "Act III" - and after the final choice it MUST include the word "Ending" or "Epilogue" - A log container with data-testid="story-log" that appends entries; each entry must have class "entry" - A restart button with data-testid="story-restart" that returns the story to Act I (fresh state) Mandatory UI skeleton (you may style it, but keep these exact data-testid values): - <button data-testid="story-start" ...>Start</button> - <div data-testid="story-act">Act I</div> (update text exactly as acts change) - <div data-testid="story-log"></div> (append <div class="entry">...</div>) - For each choice: <button data-testid="story-choice" ...>...</button> - After an ending is shown: <button data-testid="story-restart" ...>Restart</button> Output only a complete HTML document starting with <!DOCTYPE html> and including closing </body></html> tags.
Prompt SHA-256
b26085287f4bb9c7c8f6396f6473f7d419d6177241e3a33ce55a5a1ddcd67804
Claims
- Qwen artifact is preserved from the first-blog public demo regeneration flow.
- Qwen artifact was regenerated/iterated and acceptance-tested before publishing; it is not a literal one-shot controlled result.
- GPT-5.2 receives the recovered user prompt, temperature, and completion budget once; output is retained without repair.