ÆAAI
← All models
GPT54OpenAI · GPT-5

GPT-5.4

OpenAI's flagship — 77.3% on DualEntry, held

Accounting overall

77.3%

Input / Output

$2.50 / $15 per MTok

Context

1M

Speed

~90 tok/s

Released

2026-03

Cutoff

2025-12

GDPVal-AA Elo

1674

01Accounting Task Breakdown

Eight accounting-task categories borrowed from DualEntry's 101-task benchmark. Measured where published, synthesized from adjacent benchmarks otherwise.

Transaction Class.
88.0
Journal Entry
88.0
Accounts Payable
82.0
Accounts Receivable
80.0
Bank Reconciliation
76.0
Financial Reporting
60.0
Month-End Close
48.0
Accounting Knowledge
85.0
02Research

GPT-5.4 held the top spot on DualEntry's accounting benchmark for most of Q1 2026 before Opus 4.7 narrowly surpassed it. At 77.3% overall, the difference on raw accounting capability is about 2 points — but GPT-5.4 has a meaningful cost advantage: standard pricing is $2.50 / $15 per MTok (up to 272K context), roughly half Opus 4.7's blended rate.

For most agentic accounting deployments today, GPT-5.4 is the more economically rational choice. The capability ceiling is marginally lower, but the cost delta is substantial across the kind of high-volume invoice / reconciliation workloads that agentic accounting tools actually run.

Watch for: tiered pricing doubles the input cost above 272K context, which matters for tools processing long historical ledgers or multi-year financial statements in a single call. The 1M context is available but not at flat pricing.

Citations