FOR PAGEMOTOR FORUM MEMBERS
Same brief, three AI models
One job: make a launch film for justaskpage.com from its home page, using the launch-video skill. Made in Claude Code with Claude Opus 5.5, and in OpenAI's Codex with GPT-6.1 Sol and GPT-6 Astra, using the Codex edition of the same skill.
Same rules for every film. Every word on screen comes from the page, the type is drawn in code, and the music, sound effects and voice come from ElevenLabs. Each model chose its own storyboard and timing. Watch with sound: no AI has judged the voices.
Round one: not a level race
The Claude film had a full creative brief and my review notes along the way, but it was the skill's very first run, before the lessons of seven more launch films went into it. The two Codex films were each given a one-line request on today's version, 1.3.0, and left alone.
Claude Code, Claude Opus 5.5
Also on YouTube: youtu.be/pc3cLCw167E
The skill's first ever run, on 1 October 2026, made over a long session in which I reviewed it and asked for changes: the voice was re-done after it sounded American, and one label was reworded.
Length 54 s, loudness −14.7 LUFS, claims: 35 of 35 on-screen lines word for word (its own log).
OpenAI Codex, GPT-6.1 Sol (high effort)
Also on YouTube: youtu.be/hiwuVDxY5FA
Unattended, one-line request: 39 minutes, about 1,460 ElevenLabs credits, nearly all on the music.
Length 48 s, loudness −14.3 LUFS, claims: 0 lines missing (the skill's claims check).
OpenAI Codex, GPT-6 Astra (high effort)
Also on YouTube: youtu.be/nYyZb3lERkY
Unattended, one-line request: 18 minutes, about 1,460 ElevenLabs credits, nearly all on the music.
Length 48 s, loudness −14.0 LUFS, claims: 0 lines missing (the skill's claims check).
Round two: the same brief and notes for everyone
Then a fairer race. All three got the full creative brief the first Claude film started from, plus my three review notes up front: a British English female voice, ElevenLabs' Isabel on every line, and "An example interview" on any sample transcript. Same skill version, 1.3.0, all unattended from start to finish on 3 October 2026, with no notes during the run.
Claude Code, Claude Opus 5.5 (from scratch)
Unattended: 37 minutes, about 2,720 ElevenLabs credits (music about 1,680, voice about 970). Told not to look at the earlier Just Ask Page film. It voiced Isabel on ElevenLabs' older voice model, because the skill notes v4 gave her a slight American r.
Length 56 s, loudness −14.3 LUFS, claims: 0 lines missing.
OpenAI Codex, GPT-6.1 Sol (high effort)
Unattended: 50 minutes, about 2,700 ElevenLabs credits.
Length 52.8 s, loudness −13.9 LUFS, claims: 0 lines missing.
OpenAI Codex, GPT-6 Astra (high effort)
Unattended: 29 minutes, about 2,100 ElevenLabs credits.
Length 50 s, loudness −14.2 LUFS, claims: 0 lines missing.
The original Claude film, fixed on today's skill
The first Claude film again, kept as it was but with today's rules applied: Page's question typed into the empty speech bubble, a real pen drawing the marks, larger type in the interview with one exchange on screen at a time, "An example interview" on the transcript, the end card moved onto the music's final chord, and Isabel re-voiced on ElevenLabs v4. Then a 9:16 version for Reels, TikTok and Shorts, re-laid out rather than cropped.
Claude Opus 5.5, the original fixed (16:9)
Same shots, music and pace as the original, with the fixes above.
Length 54 s, loudness −14.5 LUFS on this copy, claims: 0 lines missing.
Claude Opus 5.5, the original fixed (9:16)
The same film re-laid out for a phone screen: same timing, same sound.
Length 54 s, 1080x1920, loudness −14.5 LUFS on this copy.
Claude Opus 5.5, the original fixed and narrated (v3)
The fixed film again, now narrated start to finish like the GPT films, in the round-two Isabel voice (ElevenLabs multilingual_v2): eleven spoken lines instead of three, every one taken word for word from the page. The music runs two bars longer to give the voice room. Made in answer to the point that the GPT films had better spoken audio.
Length 58.8 s, loudness −14.0 LUFS on this copy, claims: 0 lines missing (on-screen and spoken).
Every film is encoded the same way for this page (H.264, 1080p, or 1080x1920 for the vertical one). Loudness was measured on these copies; times and credits come from each run's own records.