ChatGPT’s model naming has always felt like it was optimized for changelogs, not conversation. That rang especially true this weekend when Tibo Sottiaux joked about an alternate timeline where “GPT-5.4-Pro” shipped with a more product-like name—something like “Fable”, taking a pun at the recently announced Claude Mythos model. The more concrete part: Sottiaux said “5.4-Pro is available” in a post pointing to a ChatGPT link, suggesting access has started showing up in the app experience discussed on X.
A quiet availability note—and a loud set of practical questions
The thread quickly drifted from naming into the problems that actually shape day-to-day AI-assisted coding.
Connectors and workflow integration: GitHub gets a mention
Beyond raw access, a few comments pointed at missing integrations:
- One question asked whether the 5.4 Pro model works with the GitHub connector.
- Another reply claimed it can’t be used in Codex yet.
Mythos Preview vs GPT-5.4 Pro: a small benchmark overlap, clear pricing contrast
Separately, additional context around GPT-5.4 Pro highlights a limited but interesting apples-to-apples comparison versus Anthropic’s Mythos Preview, based only on the four benchmarks both parties report:
- GPQA Diamond: Mythos 94.5 vs 5.4 Pro 94.4 (≈ tie)
- Humanity’s Last Exam (no tools): Mythos 56.8 vs 5.4 Pro 42.7
- Humanity’s Last Exam (with tools): Mythos 64.7 vs 5.4 Pro 58.7
- BrowseComp: Mythos 86.9 vs 5.4 Pro 89.3
Pricing data in that same context shows GPT-5.4 Pro listed at $30/MTok input and $180/MTok output, with long-context priced at 2× ($60 in / $270 out). Mythos Preview is listed at $25/MTok input and $125/MTok output, and long context at the same price—though access is described as “Glasswing only,” while GPT-5.4 Pro is noted as Public API.
