Skip to content

GPT-5.4 Pro

Released 2026-03 · thinking · 272K tokens · 2 benchmarks

Editorial notes

Lanzado 5 marzo 2026 (junto a GPT-5.4 base). Variante xhigh con 5-10 min de reasoning effort en decisiones criticas. Solo en Pro/Enterprise. Primer modelo OpenAI en exceder baseline humano (72.4%) en OSWorld-Verified. Pricing $30/$180 per MTok.

Spec sheet

Company
OpenAI
Country
US
Type
thinking
Release
2026-03
Context
272K tokens
Slug
gpt-5-4-pro

Benchmarks (2)

Cite this model
BibTeX · APA

BibTeX

@misc{frontier-gpt-5-4-pro,
  title  = {GPT-5.4 Pro},
  author = {{OpenAI}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
  url    = {https://frontierbenchmarks.com/models/gpt-5-4-pro}
}

APA

OpenAI (2026). GPT-5.4 Pro [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/gpt-5-4-pro

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.