We measured what PowerPoint work costs when a large language model does it natively, versus routing the same action through auxi. Across 92 auxi features in 9 categories, auxi used about 89% fewer tokens per action.

Request the full report See the auxi connector

Measured on Claude Sonnet 4.6 API pricing with prompt caching.

~89%

Fewer tokens per action

Average across all features measured.

$0.0012

Per action with auxi

Versus $0.0103 when done natively by the LLM.

96%

Saved on table operations

Up to 96%, the heaviest work to do natively.

Features analyzed

Across 9 categories of PowerPoint work.

Why the gap exists

AI drafts fast. Then you pay twice to finish.

General-purpose AI tools rewrite slide code from scratch for every change, and still hand back slides that ignore your template and chart standards. You pay in tokens to generate them, then in hours to fix them.

What auxi changes

Your AI decides what the slide says. auxi builds it.

With the auxi connector or Darwin, the model calls a ready-made PowerPoint action instead of writing it out. Fewer tokens per action, and output that lands on your template the first time.

What we measured

Same action, two ways.

For each feature, we compared the cost of the model doing the work natively with the cost of the same request routed through auxi.

Done natively by the LLM

Routed through auxi

Cost per action

$0.0103

$0.0012

Tokens per action

Baseline

About 89% fewer

Table operations

Most expensive category

Up to 96% saved

Output

Approximates your template

Built on your master and placeholders

Scope: 92 auxi features across 9 categories, priced on Claude Sonnet 4.6 API pricing with prompt caching. Results on other models will differ. The full methodology will be published in the complete report.

At scale

What it means for a firm.

An illustrative example, using the per-action costs above.

100,000PowerPoint actions a month

$1,030done natively by the LLM

$120routed through auxi

Illustrative only: 100,000 actions × $0.0103 versus $0.0012 per action. Token cost is one part of the saving; the larger part is the hours not spent fixing slides. On auxi Enterprise, AI is one price per seat.

FAQ

About the token savings report.

See auxi Enterprise

Which model did you measure?

Claude Sonnet 4.6, using its API pricing with prompt caching.

Does this apply to ChatGPT or Copilot?

The same principle applies, because auxi replaces generated slide code with ready-made actions. Exact savings depend on the model and its pricing.

How do I get these savings?

Use Darwin in PowerPoint or on the web, or connect your AI assistant to auxi through the auxi connector (MCP).

Where is the full methodology?

The complete report is being finalized. Request it and we will send it when it is published.

See your own deck, on your brand, in 30 minutes.

Book a live demo with our team, or try auxi Pro free for 14 days.

Book a demo Start free trial

Free 30-minute demo · SOC 2 · GDPR · PowerPoint 2016+ on Windows and macOS