GPT 5.2: inspect the real-code result alongside the launch claims

Read the launch material as attributed context

The video explains changes from GPT 5.1 to GPT 5.2, discusses GDP Val and covers coding, mathematics and reasoning benchmarks. The benchmark section begins at 4:01, followed by playground demonstrations at 7:04. These are reported launch comparisons, not an independent benchmark study performed for this companion.

The description also mentions Notion, Databricks and Shopify as company use cases. Those examples provide context for data, document and agentic-coding work; they do not establish that the website redesign later in the video achieves the same results.

1. Identify the task you care about before interpreting a benchmark.
2. Keep the reported metric attached to its stated model and comparison.
3. Separate a playground example from work on the real repository.

Follow the actual coding route

The real-code example uses GPT 5.2 through OpenRouter in RooCode inside VS Code, redesigning Luke’s Heavenly Bible Verses website. It begins at 10:56, after vision, tooling and use cases at 9:57. Keep the provider, coding interface and model selection visible when following the workflow.

Inspect the starting site and the intended change. A redesign should still preserve the site’s information and important visitor actions; a new appearance alone does not establish improved conversion or correct implementation.

1. Confirm the GPT 5.2 route through OpenRouter and RooCode shown in the recording.
2. Compare the generated interface with the original site and brief.
3. Inspect the changed page’s behaviour and record any follow-up correction needed.

Judge the test and sponsor segment separately

Thanks to Skywork for sponsoring this video. Its original destination is https://skywork.ai/p/E7AkCS and the description advertises code ailuke20 for 20% off. Those are historical offer details, not a current verified entitlement. The sponsor segment covers turning video into guides, slides, podcasts and web pages.

Luke’s reaction to the redesign is a judgement about that run. Keep it with the visible UI changes and the actual route used. The description does not supply a controlled coding benchmark or prove a universal improvement over every earlier model.

The takeaway

Use launch figures to frame questions, then assess the real-code example through its inputs, changes and working page.