Do you know the true cost of your AI solution?

Technology

With AI adoption accelerating across Australian accounting firms, practices are rapidly deploying advanced Large Language Models (LLMs) to help with firm workflows. However, the recent release of next-generation models by frontier platforms provides an eye-opening insight into future cost.

08 July 2026 By ElfWorks.ai 5 minutes read
Share this article on:

These costs pose a significant commercial risk, not just through the unpredictable, compounding operational overhead of high-volume token consumption, but also through the hidden cost of client data security and regulatory compliance.

We sat down with Ian Youngman, Co-Founder and Managing Director of Elfworks, to unpack the economics of modern AI models and analyse the expenses many firms overlook when integrating AI into form workflows

What is the true cost of frontier reasoning models?

Anthropic’s latest releases show the direction these platforms are heading. Fable 5 comes with a 100% price increase on their prior reasoning model with Terms that enable Anthropic to analyse every single query and store your information for 30 days for “Safety”. This has massive implications both financially and from a privacy perspective and whilst this is currently limited to Fable 5, I’m yet to see any of these companies put the genie back in the bottle. Youngman says. 

How can you bring this to life for firms to understand commercially. 

A simple example: A mid-sized accounting firm configures task to summarise a trust deed that 12 pages and roughly 6000 words. 

Based on the latest Claude pricing1 and their new tokeniser (that produces approximately 30% more tokens for the same text) the cost with Anthropic models would be

  1. Claude Fable 5: $0.36 per letter.
  2. Claude Sonnet 5: $0.12 per letter

To put this in perspective, Elfworks leveraging a combination of sovereign and frontier models does this for $0.04 per letter. That’s 9x cheaper than Fable and 3x as cheap than Sonnet for a very simple task. Extrapolate that across your firm and it becomes material. 

What happens to the data? The other cost…

"Accounting firms face two major issues when integrating AI: the unpredictability of direct consumption costs, which we’ve covered and the strict compliance demands of keeping client information secure and sovereign," says Youngman.

When client information – such as the details contained within a client Trust Deed or Engagement letter – are shared with an overseas based LLM, you need to carefully check the terms and conditions regarding use of data.

For example, under the Claude Fable terms and conditions2, every query is read for ‘safety’ and data is stored for 30 days” says Youngman. This is on top of the US Cloud Act that compel US companies to provide all data on request.   

If you have not obtained client consent to use their data based on frontier model Terms and Conditions, firms can run into problems with the Code of Professional Practice and the Australian Privacy Act. 

A local solution to cost and compliance

With over 700 accounting practices utilizing Elfworks - including 15 of Australia's Top 50 firms - the platform has established a leader in this space with a predictable commercial framework alongside its robust data protection protocols.

Instead of exposing your firm to the commercial volatility of fluctuating USD-based API usage or the strict requirements of cross-border data disclosure under APP 8 and the US Cloud Act, Elfworks streamlines both cost and security

  • Predictable credit system: Budget your software spend with confidence. Under the Elfworks credit model (1 credit = 100,000 tokens), firms know exactly what each client structure review, tax plan, or trust resolution draft will cost in advance, with no surprise bills.
  • Sovereign Australian environment: By utilising Australian-hosted infrastructure, your data stays within local jurisdictions, bypassing the risks of foreign legal compulsion and eliminating the need for complex, explicit client consent for offshore data processing.
  • Built-in anonymisation: Sensitive client details are systematically encrypted before processing and securely decrypted locally - protecting client privacy while leveraging the deep reasoning capabilities of leading global models.

The bottom line for Australian practices

As professional standards, profit structures, and technology requirements continue to tighten, managing your firm’s operational expenses is just as critical as managing your compliance.

Before you run your next high-volume batch of client documents through a direct public cloud connection, ask yourself: Do you really know what your AI solution is costing you?

See how Elfworks'  flexible plans, from a free tier through to unlimited firm-wide access, at elfworks.ai/pricing.

For Claude pricing see here: https://platform.claude.com/docs/en/about-claude/pricing

Terms and Conditions: https://www.anthropic.com/claude/fable

Accountants DailyWant to see more stories from trusted news sources?
Make Accountants Daily a preferred news source on Google.
Tags: