GPT-5.4

Name: GPT-5.4
Author: OpenAI

OpenAI · 2026-03-05

GPT-5.4 is OpenAI's frontier model for complex professional work, combining advanced reasoning, coding, and agentic workflows with configurable reasoning effort from none through xhigh. It is OpenAI's first general-purpose model with native computer-use capabilities, incorporates the coding capabilities of GPT-5.3-Codex, and reduces factual errors relative to GPT-5.2.

Benchmark scores

Benchmark	Score
A-Fantasia	42.3
A-Fantasia (Valid Attempts with Retries)	42.3
A-Fantasia Backwards Spelling	26.0
A-Fantasia Chess	33.0
A-Fantasia Chess (Valid Attempts with Retries)	33.0
A-Fantasia Cube (Valid Attempts with Retries)	68.0
A-Fantasia Cube Rotation	68.0
A-Fantasia Spelling (Valid Attempts with Retries)	26.0
AA-LCR	74.0
AA-LCR v1.1	76.7
AA-Omniscience Index	5.8
AdvancedIF	81.0
AdvancedIF (rubric mean score)	95.0
Agent Red Teaming k=100 (CAIS)	59.7
Agent4Sci mean relative improvement	8.1
Agents' Last Exam (ALE)	20.5
Agents' Last Exam ALE-CLI Pass Rate	22.5
Agents' Last Exam ALE-CLI Score	41.8
Agents' Last Exam Full-Spectrum Pass Rate	19.4
Agents' Last Exam Full-Spectrum Score	34.3
Agents' Last Exam Last-Exam Pass Rate	0.0
Agents' Last Exam Last-Exam Score	8.1
Agents' Last Exam Near-Term Pass Rate	33.6
Agents' Last Exam Near-Term Score	57.8
Agents' Last Exam Overall Score	37.3
Agents' Last Exam Unlicensed Full-Spectrum Pass Rate	19.3
Agents' Last Exam Unlicensed Full-Spectrum Score	34.9
Agents' Last Exam Unlicensed Last-Exam Pass Rate	0.0
Agents' Last Exam Unlicensed Last-Exam Score	10.3
Agents' Last Exam Unlicensed Near-Term Pass Rate	33.6