‹ PublicAI Index
The LLM benchmark aggregator.
GPT-5.4
OpenAI
Strongest in Reasoning (#15 of 195, on 3 of its 5 boards), weakest in Fairness (#212 of 292). Above par in 27 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of Gemini 3.8 Flash and MiMo V2.6 Pro.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Reasoning59.1−6.3#15/1953/5
Mathematics59.2−4.6#15/1522/2
Science60.9−2.6#21/1221/1
Reasoning56.9−12.1#40/1471/3
Professional55−7.4#30/1941/2
Legal57.2−8.7#24/1731/1
Finance57.1−5.3#29/1621/1
Medical52.3−14.2#73/1491/1
Human preference63.9−4.9#30/3541/1
Human preference63.9−4.9#30/3541/1
Core abilities54.9−10.4#46/1992/4
Language56−10.6#23/621/1
Instruction following53.9−19.9#27/621/1
General intelligence55.4−14#48/1781/3
Safety55.3−6.1#47/3292/3
Secure code63.4−2.8#24/2701/1
Factual grounding57.5−13.2#32/1041/1
Harm refusal56.1−5.4#48/2921/2
Jailbreak resistance57.8−7.2#70/2681/1
Toxicity avoidance55.1−3.9#102/2681/1
Fairness44−26#212/2921/2
Coding53−13.2#60/1792/6
Agentic coding53.1−13.8#57/1732/5
Knowledge53−9.5#64/1972/3
Academic knowledge57.8−6.1#27/1231/1
Document parsing52.3−11#42/831/1
Agents53.5−12.9#74/2981/6
Knowledge work57−13.7#57/2081/2
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for GPT-5.4, left for the other.
§ 3 · Sources
Where the numbers come from
9 publications, 39 figures. Every one links to the page it was read from.
LMArena Text 1475
GDPval-AA 1256
ARC-AGI-2 74%
LiveBench 1.0 78LiveBench 1.0 · Reasoning 88.1LiveBench 1.0 · Coding 77.5LiveBench 1.0 · Agentic Coding 53.8LiveBench 1.0 · Mathematics 94.1LiveBench 1.0 · Data Analysis 79.3LiveBench 1.0 · Language 82.6LiveBench 1.0 · Instruction Following 70.2
Kagi LLM Benchmark 63.8%
Vals · CaseLaw 63.77%Vals · LegalBench 86.04%Vals · CorpFin 65.27%Vals · TaxEval 73.96%Vals · MortgageTax 68.32%Vals · MedQA 96.09%Vals · MedCode 41.29%Vals · MedScribe 77.55%Vals · SWE-bench Verified 78.2% (Mini-SWE-agent)Vals · Vibe Code Bench 67.42% (OpenHands)Vals · Code Migration 34.98%Vals · GPQA Diamond 91.67%Vals · MMLU Pro 87.48%Vals · AIME 96.67%
ParseBench · Overall 61.72%ParseBench · Tables 83.64%ParseBench · Charts 65.57%ParseBench · Content faithfulness 86.78%ParseBench · Semantic formatting 55.67%ParseBench · Visual grounding 16.95%
Enkrypt · Jailbreak risk 6.3%Enkrypt · Harmful content risk 0%Enkrypt · CBRN risk 16.7%Enkrypt · Toxicity risk 2.7%Enkrypt · Bias risk 87.1%Enkrypt · Insecure code risk 5.8%
Vectara · Factual consistency 93%
Badge
[](https://publicai.io/model-index/m/gpt-5-4)