Back to the index
OpenAI · GPT
No. 09

GPT-5.4

The affordable workhorse: near-flagship reliability at half the price.

BalancedClosed weightsText, Image, Pdf
Field notes
Context window
1.1Mtokens
Max output
128Ktokens
Input price
$2.50/ 1M tokens
Output price
$15/ 1M tokens
Cached input
$0.25/ 1M tokens
Modalities
Text, Image, Pdf
Knowledge cutoff
Oct 2025
Released
Mar 2026
Approx. speed
Overview

What this model is

GPT-5.4 is OpenAI's balanced workhorse, delivering near-flagship reasoning and coding at exactly half the per-token cost of GPT-5.5. It was the flagship for about six weeks before GPT-5.5 launched and retains genuinely reliable agentic behavior — an OpenAI-grade default for most production workloads that want strong intelligence without flagship pricing. Roughly 1M-token context with full multimodal input.

Strengths

  • Strong, reliable all-round reasoning and coding
  • Half the price of the flagship
  • Roughly 1M-token context window
  • Multimodal text, image, and PDF input
  • Good balance of speed and quality

Trade-offs

  • Trails GPT-5.5 on the hardest tasks
  • Pricier than mini/nano for bulk work
  • Long-context premium applies above 272K tokens

Best for

  • General-purpose production assistants
  • Everyday coding and refactoring
  • Document analysis and summarization
  • Agentic tasks on a budget
  • RAG over large corpora

Not ideal for

  • Absolute frontier reasoning (use GPT-5.5 Pro)
  • Ultra-cheap high-volume classification (use nano)
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning90Coding90Math90Writing88Knowledge90Speed62Agentic88Vision87Multilingual88Long Context88
Reasoning
90
Coding
90
Math
90
Writing
88
Knowledge
90
Speed
62
Agentic
88
Vision
87
Multilingual
88
Long Context
88
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

Independent index
Artificial Analysis Intelligence Index
Composite of ~9–10 independent evals · Artificial Analysis
51/ 100
BenchmarkResult
Artificial Analysis Intelligence Indexindependent
GeneralArtificial Analysis
51%
SWE-bench Verifiedindependent
CodingSWE-bench
76.9%
SWE-bench Proindependent
CodingSWE-bench Pro
59.1%
τ-benchindependent
Agenticτ-bench
78.3%
GPQA Diamondindependent
ReasoningIndependent aggregators
88.4%
LiveBenchindependent
GeneralLiveBench
80.3%
Terminal-Bench Hardindependent
AgenticArtificial Analysis
57.6%
DeepSWEindependent
CodingDatacurve
56%
LMArena Eloindependent
GeneralLMArena
1495Elo
Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.