6.1 SOL est là : mais il n'est pas à la hauteur (tu vas être heureux)

6.1 SOL est là : mais il n'est pas à la hauteur (tu vas être heureux)

6.1 SOL is here: but it doesn't live up to the hype (you're going to be happy)

Title history
  • 6.1 SOL est là : mais il n'est pas à la hauteur (tu vas être heureux) — observed on 20261031-2319 UTC (YouTube)

by Melvynx · View the channel on PSI

Formal & Physical Sciences Computing & Cybersecurity UMComputer programmingUMXProgramming and scripting languages: general
🎙 Melvynx 👥 67K 📅 September 30, 2026 ⏱ 28 min 👁 15K 📄 expert opinion 🧭 2026-10-31
Available in: English (current) Français

Keywords

GPT-6.1 SolOpus 5.5benchmarkCodex CLIcost per task

Summary

In this video, Melvynx reviews OpenAI’s latest model, GPT 6.1 Sol, released after the DevDay event. He compares it extensively with Anthropic’s Opus 5.5, using official benchmarks and his own hands-on tests. He notes that while GPT 6.1 shows improvement over its predecessor (from score 48 to 52), it still lags behind Opus 5.5 in subjective quality and design. He highlights the cost advantage: GPT 6.1 is significantly cheaper per task, with a cost per unit that is 30% lower than GPT 6, and much cheaper than Opus 5.5. He discusses new features like Codex CLI improvements, the Dot agent, Astra ultra-fast, and the new $500 plan, which he criticizes due to reduced weekly limits on lower tiers. He runs several practical tests: generating a responsive app, creating a map-based game, and video generation. In all tests, Opus 5.5 outperforms GPT 6.1 in visual quality and detail, despite similar intelligence scores. He also shares a stick duel game he built with Opus 5.5, showcasing its capability. He concludes that GPT 6.1 is not up to par with Opus 5.5, though it is cheaper. He warns about subscription changes and advises viewers to consider their own needs.

196 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial value through direct A/B testing of two leading AI models, offering concrete examples of performance differences beyond simple benchmark scores. The creator’s argumentation is structured: he first presents official benchmarks, then analyzes cost per intelligence, and finally demonstrates practical outcomes. He acknowledges his bias towards Opus but supports his claims with multiple test cases (responsive app, game, video). The cost analysis is particularly insightful, showing that GPT 6.1 is cheaper but often yields inferior results, which raises questions about value for money. However, the argument is largely subjective, relying on personal aesthetic judgments and qualitative comparisons. He also critiques OpenAI’s pricing and feature changes, providing a balanced view that is not purely promotional.

Scientific Rigor, Source Quality, Title Accuracy

The video relies on official OpenAI announcements and benchmarks, but these are not independently verified. The creator’s own tests are transparently described, though they are not reproducible from the video alone. He mentions specific models and features, but provides no external sources beyond his own configuration links. The title accurately reflects the content, which is clearly an opinionated review. The lack of citations to peer-reviewed studies or official documentation weakens the scientific rigor, but the practical demonstrations add empirical weight. The video’s subtitle ‘you’ll be happy’ suggests a partisan tone, but the creator insists on impartiality. Overall, the sources are limited to the creator’s own claims and OpenAI’s promotional materials, limiting the depth of verification.

247 words

Title / Content Match

The title accurately reflects the content: the video argues that GPT 6.1 Sol is not nearly as good as Opus 5.5, and the creator expresses satisfaction in saying so.

Quality & Reliability

7/10

The video provides a detailed, hands-on comparison of GPT 6.1 Sol and Opus 5.5, including benchmarks, cost analysis, and practical tests. The creator shares personal opinions and clear caveats about his bias, but the technical data and ephemeral tests add credibility. However, the evaluation is subjective and lacks external verification of benchmark results.

Chapters

Cited Sources

  • Stick Duel game built by Melvynx with Opus 5.5 — Highlighted at 22:37 as a showcase of Opus 5.5's game development capabilities.
  • Free AI configuration setup (Claude Code + Codex) — Mentioned at 11:48 as the creator's recommended free setup for AI tools.

Concurring Sources

External References

Contribution & Novelties

The video offers a fresh, critical perspective on a major AI release, going beyond hype to evaluate practical performance and cost-effectiveness. It introduces the concept of ‘cost per intelligence’ and compares models across multiple tasks, revealing that cheaper models can be more economical but may produce inferior output. The creator’s hands-on tests with game and app development provide tangible evidence of differences that benchmark scores alone miss.

Pour aller plus loin :

  • AI benchmarking — Provides context on how AI models are evaluated and the limitations of standard benchmarks.
  • Cost per token in LLMs — Relevant for understanding the cost dynamics discussed in the video.
  • Anthropic’s Claude models — Background on Opus 5.5, the competitor model. (URL is likely correct but not verified)
  • OpenAI Codex — For understanding the Codex tool ecosystem.

132 words

Radar Profile

The radar profile shows a well-rounded video with strong information quantity and technical level, but moderate reliability and quality due to subjective opinions and lack of external validation. The high technical score reflects detailed coverage of AI model features, while the lower reliability score stems from reliance on personal testing and the creator's admitted bias.

Reliability 6/10