benchmarks/robustness-audit.js: hardcodes invalid model gpt-5.4-mini as default

[nanaubusiness]

Line 23 hardcodes the default model:\n\n\n\n is not a valid OpenAI model ID (the real model series is , , etc.). When AUDIT_MODEL is not set, the script silently sends requests that will fail with a 400 error. Should use a valid default model ID or require AUDIT_MODEL to be set explicitly.

[nanaubusiness]

Already fixed in PR #402.

[nanaubusiness]

Reopening — corresponding PR is under review.

[DietrichGebert]

Closing as invalid. gpt-5.4-mini is a valid, current OpenAI model, not a bad ID. Verified against OpenAI docs: gpt-5.4-mini is a callable alias (its pinned snapshot is gpt-5.4-mini-2026-03-17), described as their strongest mini model. https://developers.openai.com/api/docs/models/gpt-5.4-mini

The premise here (that gpt-5.4-mini is not a valid model and requests fail with a 400) is incorrect. It looks like a knowledge-cutoff artifact: the GPT-5.4 series shipped March 2026, so from an earlier vantage point the valid model reads as a typo and gpt-4o-mini as the obvious replacement. The benchmark also already logs real results on gpt-5.4-mini (see the results comment in robustness-audit.js), so it is the deliberately chosen, working default. No change needed.