GPT-5.4

OpenAI · Closed weights · Released Mar 5, 2026 · $2.50 in · $15 out per 1M tokens

51Overall · rank 18

Score by job

66 · rank 12 · 4 of 5 benchmarks

Benchmark	Result	Variant	z
DeepSWE	51.8%	gpt-5.4-2026-03-05_xhigh	-0.34
Terminal-Bench	81.8%	gpt-5.4-2026-03-05_unknown	1.04
SWE-bench Verified	76.9%	gpt-5.4-2026-03-05_high	0.15
LMArena Coding	1496 rating	gpt-5.4-high	0.24

43 · rank 9 · 2 of 4 benchmarks

Benchmark	Result	Variant	z
APEX-Agents	52.4%	gpt-5.4-2026-03-05_unknown	0.01
METR Time Horizons	5.7 h	gpt-5.4-2026-03-05_unknown	-0.15

59 · rank 14 · 3 of 4 benchmarks

Benchmark	Result	Variant	z
GPQA Diamond	93.3%	gpt-5.4-2026-03-05_xhigh	0.70
ARC-AGI-2	74.0%	gpt-5.4-2026-03-05_xhigh	0.46
Humanity’s Last Exam	36.2%	gpt-5.4-2026-03-05_xhigh	-0.19

59 · rank 13 · 5 of 5 benchmarks

Benchmark	Result	Variant	z
FrontierMath Tiers 1–3	78.6%	gpt-5.4-2026-03-05_xhigh	0.73
FrontierMath Tier 4	49.0%	gpt-5.4-2026-03-05_xhigh	0.34
OTIS Mock AIME	97.8%	gpt-5.4-2026-03-05_high	0.54
ProofBench	56.0%	gpt-5.4-2026-03-05_xhigh	0.33
LMArena Math	1490 rating	gpt-5.4-high	0.37

61 · rank 24 · 2 of 2 benchmarks

Benchmark	Result	Variant	z
LMArena Creative Writing	1440 rating	gpt-5.4-high	-0.32
LMArena Instruction Following	1470 rating	gpt-5.4-high	0.21

20 · rank 38 · 1 of 1 benchmarks

Benchmark	Result	Variant	z
LMArena WebDev	1463 rating	gpt-5.4-high (codex-harness)	-0.94