GPT-6 Luna Tested on a Real Board Deck

Fifteen cents for a board deck is remarkable. What it buys is a summary of what happened, not an explanation of why.

By Ed · Published November 5, 2026 · 4 min read

Twenty-five years starting companies, consulting, owning a multi-million-dollar P&L and leading business intelligence for large enterprises.

GPT-6 Luna built an 11-slide board review for 15 cents, the cheapest of seven models, in 8.6 minutes. It is a clean summary of where the ad money went and which regions grew. It did not look at campaign returns or at why the Southeast fell, and recommended putting more effort into the regions already growing. It found 1 of 8 story points in our checklist.

How this was tested. One prompt, “Northvale 2026 year-to-date board review, with an executive summary and recommendations”, run once per model in the Quiriz PowerPoint add-in on November 5, 2026. Northvale is a fictional US apparel company with stories planted in its data: Google and Meta ad exports and a HubSpot wholesale pipeline, through September 15, 2026. Every step ran on the model under test: plan, build (answering each slide’s question from the data), gap analysis, write and critic; see how a deck is built. Three small parts stayed fixed for every run (chart styling, question routing and a grounding check). Cost is what OpenRouter billed. Build time runs from pressing Build to the last slide written, and leaves out the plan, which waits for a person to approve it. One run per model, so treat close results as a tie. Full method on the benchmark page.

The run

Billed cost	$0.15 ($0.11 answers, 1 cent writing and critique)
Build time	8.6 minutes
Analysis calls	47, reading 1.7 million input tokens
Slides	11 content slides
Stories found	1 of 8 (see the checklist)

What it covered

A KPI row of Google spend, Meta spend, wholesale won amount and deals won.
Where the added ad money went: PMax-Retarget took $3,601 (61.2%) of Google’s rise, and Prospecting-Lookalike $3,989 (82.3%) of Meta’s.
Regional growth year over year: the Midwest up $123,750, the West $112,240, and the Southeast the only region down, by $32,100.

GPT-6 Luna deck: Prospecting-Lookalike drove 82.3% of Facebook cost growth — Where the added Meta spend went, from the GPT-6 Luna deck.

GPT-6 Luna deck: Midwest and West drove most wholesale deal growth — Wholesale growth by region, year over year, from the GPT-6 Luna deck.

What it left out

Returns. The deck shows where the ad money went but not what it earned, so it does not show that Prospecting-Lookalike was returning less than it cost by June.
Why the Southeast fell. It reports the decline but never looks at win rate or loss reasons, so it misses that the Southeast closed 4 of 20 deals, mostly lost on price.

Its recommendation

One choice, laid out as two options with what each gets and gives up: put existing sales effort on the Midwest and the West, or on Southeast recovery. It recommends the first because those regions grew most. The growing regions already win about half their deals; the Southeast, losing four in five, is the larger opportunity.

Verdict

Luna is a remarkable price for a “what happened” summary. It is not an analyst. For 15 more cents, DeepSeek V4.1 Flash found the ad problem. For 14 times the price, GPT-6.1 Sol adds careful efficiency analysis but still leaves the Southeast out.

Build a board deck from your own data

Describe the deck in PowerPoint and see the source of every figure. Free to start.

Try Quiriz free →

Frequently asked questions

Is GPT-6 Luna good enough for business reporting?

For a summary of totals and changes, yes. For analysis it is light: in this test it did not look at campaign returns or at why regions moved, and made one recommendation.

How much did the GPT-6 Luna deck cost?

15 cents billed for the whole run, 11 cents of it answering the data questions. List price is $0.10 per million input tokens and $0.50 per million output tokens.

How does GPT-6 Luna compare with DeepSeek V4.1 Flash?

DeepSeek cost twice as much, 30 cents, and found the biggest ad problem in the data, a campaign returning less than it cost. For a few cents more, DeepSeek is the better cheap model on this evidence.

What is GPT-6 Luna good at?

Short, well-defined steps: in our pipeline it routes questions to the right data for every model. A full analysis is a long chain of judgment calls, and that is where it falls behind.