I Made Opus 5.5, Fable 5.1 & GPT-6 Build the Same App (RAW RESULTS)

I Made Opus 5.5, Fable 5.1 & GPT-6 Build the Same App (RAW RESULTS)

by Pat Simmons · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Pat Simmons 👥 26K 📅 September 25, 2026 ⏱ 36 min 👁 676 📄 expert opinion 🧭 2026-09-25
Available in: English (current) Français

Keywords

AIapp generationClaude Opus 5.5Fable 5.1GPT-6 Astra

Summary

This video by Pat Simmons presents a head-to-head comparison of three leading AI models—Claude Opus 5.5, Claude Fable 5.1, and GPT-6 Astra—tasked with building the same applications from identical prompts. Three distinct builds are showcased: an animated risograph-style Moby Dick scene generator entirely in code, a launch video and website clone for Opus 5.5, and a Tony Hawk’s Pro Skater clone using Blender and Three.js. Each model run is observed live, with detailed notes on timing, costs, and practical observations. The creator uses consistent high effort settings and provides full prompts and live links in a companion blog post. Results show Opus 5.5 consistently producing the most visually detailed and faithful outputs, while GPT-6 Astra is fastest but less precise. Cost analysis, based on API-equivalent pricing from session logs, reveals significant differences, with Astra being cheaper in some builds. Overall, the video offers valuable hands-on insights into current AI coding capabilities, though cost figures are approximate and based on self-reported session logs.

162 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial practical value by conducting real-world tests with standardized prompts, yielding concrete artifacts that viewers can inspect. The creator’s methodology of using identical prompts and observing live builds adds credibility, and his willingness to share prompts and code links enhances transparency. The argumentation is coherent: he presents visual evidence, notes nuances in output quality, and gives explicit comparisons (e.g., ‘Opus clearly won that one’). However, the evaluation relies heavily on subjective visual assessment rather than objective metrics, and the cost approximations are rough estimates derived from session logs, not official billing. The creator acknowledges these limitations, which strengthens the overall argumentative honesty. The inclusion of multiple build types (creative, marketing, game) helps generalize findings, though the sample size is small and no error statistics are provided.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is moderate for a non-formal test. The creator controls for many variables: same prompts, same effort level, same deployment environment, and documents usage and costs. He also follows up on model behaviors (e.g., asking models to deploy when they forgot). Transparency is high, as he shows raw outputs and admits when results are imperfect. The main sources cited are the creator’s own blog post (containing prompts and live links) and a newsletter sign-up. No external academic papers or official documentation are referenced. The title is accurate and reflects the content, and the video makes no false claims about the models. Adequation between title and content is strong. The lack of peer-reviewed sources and the reliance on self-reported data limit the rigor, but the video appropriately positions itself as a hands-on test rather than a scientific study.

283 words

Title / Content Match

The title accurately reflects the content: a head-to-head comparison of three AI models building the same applications with minimal intervention.

Quality & Reliability

7/10

The video presents a consistent methodology, transparent about limitations and costs, but relies on subjective assessment and approximate cost calculations.

Chapters

Cited Sources

Contribution & Novelties

This video contributes a comparative evaluation of three frontier AI models in real-time app generation, offering a rare live demonstration of their creative and technical capabilities under identical conditions. The detailed cost breakdown using API-equivalent pricing from session logs provides a practical perspective for potential users, and the open availability of prompts and code links enables replication. The inclusion of varied tasks (creative, marketing, game) offers a broad assessment.

Pour aller plus loin :

  • Generative artificial intelligence — Context on AI systems that generate code, images, and more, directly relevant to the models tested.
  • Large language model — Background on the architecture and capabilities of models like Claude and GPT-6.
  • Blender (software) — The 3D creation suite used in the Tony Hawk’s Pro Skater clone build, illustrating its role in AI-assisted game development.

133 words

Radar Profile

The radar profile shows high scores for information quantity and quality, indicating a rich and well-presented comparison, while technical depth and reliability score lower due to the subjective nature of the evaluation and approximate cost estimates. Overall, it's a well-executed expert demonstration with caveats on precision.

Reliability 7/10