Claude Opus 4.5 : Le Modèle qui EXPLOSE Gemini 3 Pro ? (déjà)

Claude Opus 4.5 : Le Modèle qui EXPLOSE Gemini 3 Pro ? (déjà)

Claude Opus 4.5: The Model That Blows Away Gemini 3 Pro? (already)

Title history
  • Claude Opus 4.5 : Le Modèle qui EXPLOSE Gemini 3 Pro ? (déjà) — observed on 20261110-0326 UTC (YouTube)

by Julien Massey · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Julien Massey 👥 39K 📅 November 28, 2025 ⏱ 14 min 👁 1K 📄 expert opinion 🧭 2026-11-10
Available in: English (current) Français

Keywords

Claude Opus 4.5Gemini 3 Probenchmarkcode generationagentic AI

Summary

In this video, the creator tests Claude Opus 4.5 against Gemini 3 Pro, focusing on coding, agentic tasks, and productivity. He performs live demonstrations including creating a landing page, a pricing section, a slide deck from a script, and a dashboard from CSV data. He references benchmark results, notably SWE-bench where Opus 4.5 scores 80.9% versus 76.12% for Gemini 3 Pro, and discusses operational costs, noting that Opus 4.5 is cheaper than its predecessor. The content is primarily aimed at showing how these models can generate client deliverables for an AI agency, with practical examples and a business-oriented perspective. The creator also points out limitations such as encountering usage limits during longer tasks. Overall, the video serves as a hands-on product review rather than a rigorous scientific comparison, blending subjective impressions with some technical data.

135 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers concrete, time-stamped demonstrations that show the model’s ability to produce polished outputs from minimal prompts, which is valuable for practitioners and entrepreneurs. The argumentation is built on personal experience and visual evidence rather than systematic, quantitative evaluation. The creator acknowledges that the first attempt at a landing page was mediocre, then improved after a better prompt, which adds credibility. However, his enthusiasm and commercial incentives to promote AI agencies introduce bias. He justifies the value by showing real business use cases, but he does not provide rigorous performance metrics or controlled tests. The claims about cost and benchmark improvements are presented without detailed sourcing, weakening the overall argumentation.

Scientific Rigor, Source Quality, Title Accuracy

The video mentions SWE-bench as a serious test but does not link to the official benchmark or published results. It also discusses cost reductions without citing official pricing pages. The creator’s own business course is promoted, which introduces a conflict of interest. The title is sensationalist, using ‘EXPLOSE’ to imply dramatic superiority, but the video does indeed show Opus 4.5 outperforming Gemini 3 Pro in several tests. The content matches the title’s promise, though the claim of ’explosion’ is exaggerated. Overall, scientific rigor is limited by the absence of verifiable sources and the promotional elements.

221 words

Title / Content Match

The title is sensationalist ('EXPLOSE') and clickbaity, but the video does indeed compare Claude Opus 4.5 against Gemini 3 Pro, concluding that Opus outperforms on coding and agentic tasks. The content aligns with the title's promise, though the tone is exaggerated.

Quality & Reliability

6/10

The creator provides live demonstrations and concrete examples, but the evaluation is subjective, lacks a controlled methodology, and is interwoven with promotional content for his own AI agency course. Benchmark claims are mentioned without direct citation.

Chapters

Cited Sources

  • SWE-bench — Referenced as a serious benchmark for coding capabilities, where Opus 4.5 scores 80.9% and Gemini 3 Pro scores 76.12%.

Contribution & Novelties

The video provides a practical, hands-on comparison of Claude Opus 4.5 and Gemini 3 Pro in applied business contexts, demonstrating their ability to generate landing pages, pricing sections, presentations, and dashboards from concise prompts. It highlights the agentic nature of Opus 4.5, which executes multi-step tasks autonomously, and discusses cost efficiency relative to previous models. The original contribution lies in the demonstration of how these models can be leveraged for AI agency offerings, making the capabilities tangible for potential clients.

Pour aller plus loin :

  • Large language model — Foundational concept for understanding LLMs and their capabilities.
  • SWE-bench — The benchmark used in the video for coding performance evaluation.
  • Benchmark (computing) — General concept of performance testing and its limitations.

120 words

Radar Profile

The radar profile shows a moderate quantity of information with a fairly rich set of demonstrations, but the quality and reliability are hampered by subjective judgments and lack of rigorous sourcing. Technical depth is moderate, as the video touches on strengths and limitations without deep analysis, resulting in a balanced but not highly credible overall profile.

Reliability 5/10