GPT-5.4
OpenAI's flagship — 77.3% on DualEntry; 272K standard context, 1M context available at a premium.
Accounting overall
77.3%
Input / Output
$2.50 / $15 per MTok
Context
272K
Speed
~90 tok/s
Released
2026-03
Cutoff
2025-12
GDPVal-AA Elo
1674
Eight accounting-task categories borrowed from DualEntry's 101-task benchmark. Measured where published, synthesized from adjacent benchmarks otherwise.
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 capability gap is about 2 points — but GPT-5.4 has a meaningful cost advantage: standard pricing is $2.50 / $15 per MTok up to a 272K context, roughly half Opus 4.7's blended rate.
Context-window correction: prior editions of this profile listed the context as 1M tokens, which is technically available as an opt-in tier but carries a 2x-input / 1.5x-output surcharge. The flat-pricing standard context is 272K, which is what most production deployments actually run. For tools processing long historical ledgers or multi-year financial statements in a single call, plan for the tiered pricing.
For most agentic accounting deployments today, GPT-5.4 is the more economically rational choice. The capability ceiling is marginally lower than Opus 4.7, but the cost delta is substantial across the kind of high-volume invoice / reconciliation workloads that agentic accounting tools actually run.
Citations
- DualEntry — GPT-5.4 tops AI accounting testdualentry.com/blog/new-openai-gpt-5-4-real-accounting-workflow-test
- OpenAI pricingopenai.com/api/pricing
- NxCode — GPT-5.4 release specsnxcode.io/resources/news/gpt-5-4-release-date-features-pricing-2026
- Apiyi — GPT-5.4 272K standard context, 1M opt-in pricing thresholdhelp.apiyi.com/en/gpt-5-4-1m-context-272k-pricing-threshold-performance-guide-en.html