VerdictPal
Join

Models / OpenAI

GPT 5.4

GPT 5 · Released 2026-03-05

OpenAI's March 2026 model. AA Intelligence Index 56.8 (#5 globally). SWE-bench Verified 80.0%, Terminal-Bench 2.0 81.8% (ForgeCode). ARC-AGI-2 83.3% (Pro variant). 256K context, $7 / $28 per 1M tokens — 44% cheaper than GPT 5.5 on input.

Default for production OpenAI workloads where GPT 5.5 is overkill but GPT 5.2 is too old. The Pro variant leads ARC-AGI-2 among non-OpenAI competitors at 83.3%.

#5 of 51 on AA Index · current snapshot
51 AA Index · 18 benchmark rows · 10 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite61
Context256K
Input / 1M$7.0
Output / 1M$28
Knowledge cutoff2026-01-31
Output speed152 t/s
TTFT (default API)129.43s
Statusflagship
ChatGPTCursorPerplexityScira AI
KnowledgeReasoningCodingAgenticLong contextTool useHuman preference
Knowledge
90
Reasoning
92
Coding
80
Agentic
82
Long context
74
Tool use
84
Human preference
81

83average across 7 tested families

Editor's note
GPT 5.4 is the OpenAI model we recommend when GPT 5.5 is too expensive but the task still needs frontier coding. The ForgeCode variant of Terminal-Bench 2.0 (81.8%) is the most reproducible number we have on the model.
Benchmark placements

Where GPT 5.4 places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
AA time to first tokenPerformance#1/ 69Artificial Analysis2026-07-21
AA-LCRLong context#3/ 6474Artificial Analysis2026-07-21
DeepSWECoding#4/ 652DeepSWE leaderboard v1.12026-06-20
BFCLTool use#5/ 2084Berkeley Function Calling Leaderboard V42026-04-15
SciCodeCoding#5/ 6557Artificial Analysis2026-07-21
LiveBenchReasoning#6/ 2166LiveBench leaderboard2026-04-15
LongBench v2Long context#6/ 1954LongBench v2 leaderboard2026-04-01
MMLUKnowledge#7/ 2290OpenAI GPT-5.4 release2026-03-05
Terminal-BenchAgentic#7/ 6982Terminal-Bench 2.0 leaderboard2026-05-20
LMArena (Chatbot Arena)Human preference#8/ 1381LMArena leaderboard2026-05-25
Humanity's Last ExamReasoning#10/ 6542Artificial Analysis2026-07-21
Artificial Analysis Intelligence IndexReasoning#11/ 6751Artificial Analysis2026-07-21
SWE-bench VerifiedCoding#12/ 2380SWE-bench official leaderboard2026-05-30
GPQA DiamondReasoning#13/ 6992Artificial Analysis2026-07-21
ARC-AGIReasoning#14/ 140ARC-AGI-3 paper (arXiv 2603.24621)2026-03-20
AA output speedPerformance#17/ 69Artificial Analysis2026-07-21
IFBenchReasoning#22/ 5774Artificial Analysis2026-07-21
τ³-Bench BankingAgentic#26/ 6430Artificial Analysis2026-07-21
Family context

The three highest-scoring carded models in each capability family. Where GPT 5.4 shows up, it's highlighted.

KnowledgeGPT 5.4 · 90
  1. 1
    GPT 5.5
    91
  2. 2
    Claude Opus 4.8
    91
  3. 3
    Claude Sonnet 4.6
    91
ReasoningGPT 5.4 · 92
  1. 1
    Claude Mythos Preview
    95
  2. 2
    GPT-5.6 Sol
    94
  3. 3
    Kimi K3
    94
CodingGPT 5.4 · 80
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticGPT 5.4 · 82
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextGPT 5.4 · 74
  1. 1
    Kimi K3
    75
  2. 2
    GPT-5.1
    75
  3. 3
    GPT-5.6 Sol
    74
Tool useGPT 5.4 · 84
  1. 1
    Claude Sonnet 4.6
    89
  2. 2
    GPT 5.5
    87
  3. 3
    Claude Opus 4.8
    87
Human preferenceGPT 5.4 · 81
  1. 1
    GPT 5.5
    95
  2. 2
    Claude Opus 4.8
    92
  3. 3
    Gemini 3.5 Flash
    91
Newest receipts

Benchmark rows added to the public ledger for GPT 5.4 in the last 120 days. Older rows live in the full table below.

16 recent
BenchmarkFamilyScoreSourceDays ago
AA-LCRLong context74%Artificial Analysis2
AA output speedPerformance152 t/sArtificial Analysis2
AA time to first tokenPerformance129.43sArtificial Analysis2
Artificial Analysis Intelligence IndexReasoning51.4%Artificial Analysis2
GPQA DiamondReasoning92%Artificial Analysis2
Humanity's Last ExamReasoning41.6%Artificial Analysis2
IFBenchReasoning73.95%Artificial Analysis2
SciCodeCoding56.6%Artificial Analysis2
τ³-Bench BankingAgentic30.31%Artificial Analysis2
DeepSWECoding52%DeepSWE leaderboard v1.133
SWE-bench VerifiedCoding80%SWE-bench official leaderboard54
LMArena (Chatbot Arena)Human preference1405LMArena leaderboard59
Terminal-BenchAgentic81.8%Terminal-Bench 2.0 leaderboard64
BFCLTool use83.8%Berkeley Function Calling Leaderboard V499
LiveBenchReasoning66.4%LiveBench leaderboard99
LongBench v2Long context54.2%LongBench v2 leaderboard113
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Knowledge90
Reasoning92
Coding80
Agentic82
Long context74
Tool use84
Human preference81
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover or tab any dot for name, score, and input price.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMiniMaxxAIMetaZhipuXiaomiMeituanCohere

Full benchmark ledger

Every catalog benchmark for GPT 5.4. Scores link to the original source; gaps mean no public row exists yet.

18 sourced rows

18 of 32 catalog benchmarks have a sourced row for GPT 5.4.

Knowledge

Reasoning

Coding

Agentic

Long context

Tool use

Performance

Human preference

Changelog
  1. Released GPT 5.4 — AA Index 56.8, Terminal-Bench 2.0 81.8% (ForgeCode).
  2. GPT 5.2 retired for general availability.
Sources