J’ai remplacé Claude Code par Grok 4.5 dans Cursor

Value of the Information & Strength of the Argument

The video provides a genuine, hands‑on assessment of Grok 4.5, going beyond vendor claims. The author tests the model on a realistic development task with a well‑defined issue, giving concrete evidence of its capabilities and limitations. He explains the benchmarking pitfalls, such as the difference between model-only and model+agent evaluations, and highlights the cost savings as a major advantage. The argumentation is clear and structured: he compares official benchmarks, analyses the pricing, shows the setup process, and then demonstrates a complete workflow from planning to pull request. His personal workflow (using multiple AIs interchangeably) adds practical context. However, the evaluation is subjective and covers only one type of task (a moderately simple front‑end change); more demanding projects might reveal different results. The video would be stronger with a more rigorous methodology, e.g., comparing multiple attempts or including error cases.

Scientific Rigor, Source Quality, Title Accuracy

The video relies primarily on the official xAI blog post for benchmarks and pricing, which are vendor‑provided and therefore not independent. The author cites the blog and the Grok Build website but does not point to third‑party evaluations. The open‑source code is the author’s own project, available on GitHub, which reinforces transparency. The title accurately matches the content: it is indeed about replacing Claude Code with Grok 4.5 in Cursor. The reviewer’s own experience is consistent with the official claims, but the lack of external verification is a minor weakness. The video would benefit from citing independent benchmark analyses (e.g., from AI research groups) and addressing potential criticisms from the community. Overall, the source quality is acceptable for an opinion/test video, but not sufficient for a rigorous scientific assessment.

The title accurately reflects the video: the author tests Grok 4.5 as a direct replacement for Claude Code in the Cursor IDE.

7/10

The review is based on a real‑world test on an open‑source project, but the presented benchmarks come directly from the vendor's blog, introducing potential bias. The author is transparent about his subjective experience and usage costs.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction: Grok 4.5 released on July 16, 2026, and the video's purpose to test it in Cursor.
• 01:00Review of benchmark results from the xAI blog: DeepSWE, SWE-Marathon, Terminal-Bench 2.0, and SWE-Bench Pro.
• 03:00Discussion on speed (80 tokens/s) and token efficiency – Grok claims 4x lower token consumption than Opus 4.8.
• 04:00Pricing comparison: Grok 4.5 at $2/$6 per million tokens vs Claude Opus at $5/$25, making it up to 4x cheaper.
• 05:00Announcement of Grok Build being open‑source (written in Rust) and quick installation demo.
• 06:00Configuration and integration with existing .adjinns skills and MCP servers in the home directory.
• 08:00Testing Grok 4.5 inside Cursor: selecting the model, enabling Fast mode, and checking usage meter.
• 09:00Planning phase: presenting the GitHub issue and letting Grok generate a plan to add Tailwind style abstraction.
• 12:00Implementation and live coding – watching Grok modify files and create branches; results verified in the browser.
• 15:00Creating the pull request and final summary: expected usage cost (1% of Cursor quota) and overall satisfaction.

• Grok 4.5 announcement on xAI — Official blog post presenting the new model, benchmarks, and features.
• Grok Build — Official site for the open‑source terminal agent compatible with Grok models.
• UX DataTables repository — The open‑source Symfony bundle used for the live test; the GitHub issue resolved in the video.

• pentiminax.fr
• www.youtube.com
• www.youtube.com
• www.youtube.com
• www.youtube.com