GPT-5.4-Nano
OpenAI's fastest, cheapest GPT-5.4 variant — 75.2% on DualEntry, best speed-per-dollar.
Accounting overall
75.2%
Input / Output
$0.10 / $0.40 per MTok
Context
1M
Speed
~250 tok/s
Released
2026-03
Cutoff
2025-12
Eight accounting-task categories borrowed from DualEntry's 101-task benchmark. Measured where published, synthesized from adjacent benchmarks otherwise.
GPT-5.4-Nano is the surprising entry on the DualEntry leaderboard — at 75.2% it beats Mini despite being smaller and cheaper. For extremely high-volume real-time workloads (hundreds of transactions per second in an AP pipeline, for instance) where latency and cost dominate capability, Nano is the correct choice.
The Nano trade-off is consistency: while its DualEntry overall is strong, per-category variance tends to be higher than flagship models, meaning individual edge cases fail more often even if the aggregate number looks competitive. For workflows with a human reviewer on output, that's fine. For unattended agentic close, it isn't.
Citations
- DualEntry benchmark (GPT-5.4-Nano 75.2%)dualentry.com/blog/claude-opus-4-7-accounting-ai-benchmark-results