ÆAAI
← All models
CL47Anthropic · Claude 4

Claude Opus 4.7

Anthropic's flagship reasoning model —

Accounting overall

79.2%

Input / Output

$5.00 / $25 per MTok

Context

1M

Speed

~80 tok/s

Released

2026-03

Cutoff

2026-01

GDPVal-AA Elo

1753

01Accounting Task Breakdown

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

Transaction Class.
92.0
Journal Entry
92.0
Accounts Payable
85.0
Accounts Receivable
83.0
Bank Reconciliation
78.0
Financial Reporting
62.0
Month-End Close
50.0
Accounting Knowledge
88.0
02Research

Claude Opus 4.7 is the strongest publicly available model on agentic accounting workloads as of April 2026, per DualEntry's 101-task benchmark (79.2% overall, ahead of GPT-5.4 at 77.3%). The published category breakdown is instructive: Opus 4.7 hits 92% on structured tasks (transaction classification, journal entries) but drops to 62% on financial reporting and 50% on month-end close — the workflows that require multi-step reasoning and judgment.

Opus 4.7 is also the Artificial Analysis GDPVal-AA leader at Elo 1753, meaningfully above GPT-5.4 (1674) and Gemini 3.1 Pro (1314). For agentic accounting tools that can afford the higher token cost ($5 / $25 per MTok), Opus 4.7 is currently the default premium choice.

Limitations worth flagging: tokenization changed in 4.7, producing up to 35% more tokens than 4.6 for the same input — meaning the "unchanged" per-MTok price translates to higher real costs on structured data and code. The 128K output ceiling also matters for long financial reports, where the model may need chunking.

Citations