GPT‑5.4

GPT-5.4 with mid-response interruption and fewer factual errors

Visit GPT‑5.4 →

GPT-5.4 Thinking is an OpenAI language model offering deeper web research, improved long-context retention, and 33% fewer factual errors than its predecessor. It supports mid-response interruption, allowing users to redirect the model without restarting. Designed for users who need efficient, accurate AI assistance with reduced token consumption.

At a glance

Company
OpenAI
Pricing
usage-based
API available
Yes
Self-hostable
No
Launched
2026-03
Last verified
2026-05-11

Capabilities

long-contextweb-searchmid-response-interruptionreduced-token-usagefactual-accuracyreasoning

Categories

Alternatives

For AI agents: machine-readable markdown version of this page at /tools/gpt-5-4-5.md, or send Accept: text/markdown.