Grok 4.7: No-Hype Full Review & Testing

Grok 4.7: No-Hype Full Review & Testing

by Pat Simmons · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Pat Simmons 👥 25K 📅 September 22, 2026 ⏱ 22 min 👁 5K 📄 expert opinion 🧭 2026-09-22
Available in: English (current) Français

Keywords

Grok 4.7AI testingbenchmarkscost per taskcoding performance

Summary

In this video, Pat Simmons reviews xAI’s new model Grok 4.7, comparing it against Fable 5.1 and GPT-6 Astra across three coding tasks: cloning an Awwwards winning website, building an interactive 3D keyboard product page, and creating a Mario Kart clone. He first examines the benchmarks from the xAI press release, noting Grok 4.7’s strong cost-efficiency and competitive performance on certain knowledge work tasks, while highlighting that GPT-6 Astra was omitted from some comparisons. The practical tests reveal that Grok 4.7 produces the weakest results among the three models, particularly in the Mario Kart and 3D product page tasks, despite being significantly cheaper. The video provides detailed data on costs, build times, and token usage for each model, showing that while Grok 4.7 is much less expensive, its output quality lags behind its competitors. Simmons concludes that although Grok 4.7 is promising for cost-sensitive applications, it still requires improvements to match the coding capabilities of Fable 5.1 and GPT-6 Astra.

160 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers practical value by directly testing models on realistic coding tasks rather than relying solely on benchmark scores. The custom tests include prompts that require creativity, visual accuracy, and functional implementation, providing a more holistic view of model capabilities. The argumentation is structured and transparent, with the creator openly acknowledging limitations such as one-shot attempts and the narrow scope of tests. He also discusses the potential bias in benchmarks since Cursor, used for one benchmark, is owned by xAI. The cost analysis adds practical insight for developers considering model adoption. However, the tests are subjective and limited to three scenarios, which may not generalize to all use cases, but the cautious conclusions are well-supported by the presented evidence.

Scientific Rigor, Source Quality, Title Accuracy

The review demonstrates reasonable rigor: the creator uses official benchmarks from xAI’s press release, clearly notes the missing GPT-6 Astra data, and conducts his own experiments with detailed prompts. He mentions that all prompts and results are available in his blog post, providing transparency. The title accurately reflects the content, and the description includes links for further exploration. The sources used are the creator’s own materials and the official press release, which is appropriately referenced. The analysis is fair in acknowledging both strengths and weaknesses of Grok 4.7, without overhype, aligning with the ’no-hype’ claim in the title.

233 words

Title / Content Match

The title accurately reflects the content, which is a practical, no-hype evaluation and testing of Grok 4.7 against other models.

Quality & Reliability

7/10

The review is based on transparent practical testing with clear methodology, including benchmark analysis, cost tracking, and token usage. The creator acknowledges limitations such as limited test count and potential bias in benchmarks (Cursor owned by xAI). The information is presented with appropriate caveats, making it reasonably reliable for an informal review.

Chapters

Cited Sources

Contribution & Novelties

This video provides a hands-on, realistic comparison of Grok 4.7 against leading models, focusing on practical coding tasks and cost efficiency. It goes beyond official benchmarks by testing creativity, design accuracy, and functional implementation, offering insights into real-world applicability. The cost and token analysis helps users understand the trade-offs between performance and expense.

Pour aller plus loin :

  • Large language model — Provides background on LLMs and their capabilities, relevant to understanding the models tested.
  • Benchmark (computing) — Explains how benchmarks are used to evaluate system performance, contextualizing the video’s benchmark analysis.
  • Grok (chatbot) — Offers information on xAI’s Grok series, though may not cover version 4.7 specifically.

108 words

Radar Profile

The scores are relatively balanced, with quantity of information leading at 7, while quality, technical level, and reliability are each at 6. This suggests a moderately detailed and dependable review that is accessible but not deeply technical. The overall profile indicates a solid, practical evaluation with room for deeper analysis.

Reliability 6/10