GPT-5.4 (Non-reasoning) Benchmark Results

GPT-5.4 (Non-reasoning): benchmark results

GPT-5.4 evaluated with reasoning disabled. Provider: OpenAI. Released 2026-03-05. Access: API.

Aggregate Rating 1678 ± 8, rank #425 of 2168 rated models, from 227 benchmark results.

Benchmark	Score	Metric	Percentile
CDR-Bench - Atomic Filters	88.1	Recipe success RS@3 (%) on the atomic filter (single KEEP/DR	100
CDR-Bench - Atomic Mappers	30.3	Recipe success RS@3 (%) on the atomic mapper (single text-re	100
CDR-Bench - Order-Sensitive Mapper Recipes	13.99	Recipe success RS@3 (%) on the order-sensitive mapper-recipe	100
IndustryBench - Selection and Substitution	2.19	Final (SV) score (out of 3): mean 0-3 rubric score from a Qw	100
Khondo (PolySeq) - Page Typing	51.3	Page accuracy (%; share of pages assigned their correct admi	100
LexKairos - Effective Version Identification	59.96	Accuracy (%; 452 items: generate the statute title and versi	100
LexKairos - Event Relation Discerning	90.58	F1 score (%; 450 items: classify the temporal relation betwe	100
LexKairos - Global Event Ordering	98.4	Accuracy (%; 500 items: order the events of a case chronolog	100
MedConclusion	73.22	Semantic similarity (judge score 0-100): the model writes a	100
MedConclusion - Formality Similarity	89.8	Formality similarity (judge score 0-100): the model writes a	100
MedConclusion - Non-Contradiction	84.61	Non-contradiction rate (judge score 0-100): the model writes	100
MedConclusion - Numeric Consistency	88.24	Numeric consistency (judge score 0-100): the model writes a	100

Interactive version: theaggregate.ai/model?slug=gpt-5-4-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.