Comparison boards: English transcripts

1. Original instruction to both chats

ARQ: Add two Luna 6 XH chats to my pinned chats, and name all the Lunas from 1 to 3.

2. Initial progress

Gooog · 22 October 2026, 04:03

I’ll add two pinned chats with the Luna 6 model and XHigh reasoning. I’ll call the current Luna “luna 1” and the new ones “luna 2” and “luna 3”. I’m using OpenAI Docs to confirm the model setting.

Luna 1 · 03:55

I’ll check the pinned chats and all visible Luna chats, then pin the two appropriate ones and number them without changing the other names.

Gooog · 04:05

LUNA-3: the two new chats have been created, and the existing Luna is named “Luna 1”. I’m pinning the new ones beside it and checking the result.

Luna 1 · 04:02

I can see only one Luna chat in the visible list; I won’t guess which other two you want pinned. I’m checking whether the application provides a more complete list.

3. First final responses

Gooog · 04:06

LUNA-3: Done — pinned in order: Luna 1, Luna 2, Luna 3. The two new ones have GPT-6 Luna · XHigh and are waiting for instructions.

Created chat: Luna 2
Created chat: Luna 3

Luna 1 · 04:02

I can see one chat, “Luna 6XH” — it is already pinned in third position: open chat. I did not find the other two in the available list, so I changed nothing. Send me their titles or links and I’ll pin them and number all three “Luna 1–3”.

4. Additional model-label instruction, in Gooog

ARQ: Add the models as on the others.

Gooog · 04:06

I’ll add the designation in brackets to the names of the three Lunas, as on the other chats. I’m changing the names only, not the model settings.

Gooog · 04:08

LUNA-3: Added and checked: Luna 1 [L6XH], Luna 2 [L6XH], Luna 3 [L6XH]. L6XH = Luna 6, XHigh reasoning.

There is no corresponding instruction or step in the Luna conversation.

5. Request for an image and follow-up

ARQ to Gooog: Make me a screenshot of your response and Luna 1’s response to the first instruction about renaming.

ARQ to Luna: Make me a screenshot of your response and Gooog’s response to the first instruction about renaming.

Gooog · 04:08

I’ll check both responses and whether real screenshots are possible. I’m using the Computer Use instructions to establish whether I can capture the Codex window.

Luna 1 · 04:04

I’ll check the Gooog thread so the response concerns the right instruction. I cannot capture the native chat view in this mode; I’ll check whether I can prepare a faithful image of the responses without touching Gmail.

Gooog · 04:09

I have both responses to the identical instruction. Computer Use does not allow me to capture the native Codex view, so I’ll prepare an image of their verbatim content — clearly labelled as a comparison, not a screenshot of the application.

Luna 1 · 04:09

I’ll check the Computer Use instructions to establish whether a genuine screenshot can be made without capturing the private Gmail view. The Gooog thread also shows that the three Luna 1–3 names have already been changed; I’ll check exactly what I need to pass on.

Gooog · 04:11

REN-CMP: Both original responses below. This is a faithful presentation of the content, not a screenshot of the Codex window — the available Computer Use does not permit that. An image was attached in the original response, with a link to the full-resolution image.

Luna 1 · 04:10

Sent to Gooog [S6.1XH] — I corrected my earlier response. I checked: Luna 1, 2 and 3 are named and pinned in positions 3–5.

I cannot make a real screenshot of the native Codex window. The computer-use/SKILL.md I read prohibits automating the Codex interface in Windows apps: “Do not automate the ChatGPT desktop app UI or Codex CLI or Codex extensions within Windows apps.” Gooog is preparing an image of the content labelled as a comparison, not a screenshot.

6. Subsequent result and editorial assessments

The full comparison says the three chats were later confirmed in positions 3–5, with the [L6XH] labels. Two new chats were created with GPT-6 Luna and XHigh reasoning, while the previous chat retained its history. The names and pins were checked at 04:29 on 22 October 2026.

Gooog’s assessment: Gooog created two new chats beside the existing one, pinned and named them, and added the [L6XH] labels after a separate request. Luna 1’s first response did not create the new chats; it searched for existing ones and asked ARQ for their titles or links. Its later confirmation after Gooog’s changes is distinct from carrying out the original instruction. Gooog’s view was that the important difference was recognising the intended action. This single task does not show that Luna always fails, nor does it establish a winner on cost or speed. A clearer future instruction could explicitly say to create two new chats, name all three and pin them.

Luna 1’s assessment: The initial response reporting only one visible Luna was outdated; the later correction reported the proper state. Pinning the three chats and adding the model labels to their names are separate facts. The example suggests checking the full list before treating a rename task as complete.

The boards preserve the quoted times and make a separate distinction between quotations and later assessments. Empty cells mean no corresponding visible statement; they do not establish that tools were idle.

1. Plans

Gooog · 22 October 2026, 04:46

I’ll prepare a bar chart: 60 fifteen-minute intervals from the last 15 hours. I’m using visualize and will retrieve fresh measurements so I do not show an old range.

Luna 3 · 04:29

I’ll use the visualize skill; I’ll check whether reliable measurements are available every 15 minutes for the last 15 hours, then prepare a bar chart.

2. Data, range and preview

Gooog · 04:48

I have data through 04:48. The bars will show the approximate reduction in extra credits for the whole account — not usage by a single chat.

Gooog · 04:50

The chart covers 21 October 13:48 to 22 October 04:48. The data are ready; I’m still checking why the preview did not load the chart.

Gooog · 04:52

The preview still does not launch. The data and totals themselves are correct; I’ll check the display error without fetching the history again.

No separate Luna 3 messages were visible for these stages. That is not evidence that tools were idle.

3. Final responses

Gooog · 04:54

CRED-15H: 60 bars of 15 minutes each, data through 04:48. Approximate reduction in extra credits for the whole account.

A chart was embedded in the original response; its contents are separate from this text quotation.

Luna 3 · 04:32

I cannot draw usage reliably: there are no 15-minute samples for the last 15 hours. The limits reading shows the current balance, and the previous reading confirms the absence of a reliable charge history (CRED-10H-ROOT-READBACK.json). I won’t make bars up.

Your next step: provide a log of actual balances or charges for this window. Then I’ll prepare a chart with 60 bars.

4. Separate evidence and editorial assessment

Gooog delivered an approximation for 60 intervals. The cited receipt covered 1,006 account-balance observations and an estimated net reduction of 10,480.41 extra credits across the account. This was not an exact billing log, the cost of a particular chat or usage including the subscription allowance.

Luna 3 did not produce a chart and declined to invent charges without reliable history. Its conclusion drew partly on an earlier report that did not establish the absence of newer balance observations. The chats did not have identical inputs or memory, so the difference cannot be attributed solely to the model.

In the editorial assessment, a stronger model could prepare a complete prompt for Luna and identify the data source. First specify the measure: an approximate balance reduction or exact charges. The user reviews the result; savings from this division have not been measured.

The whole turns took 8 min 16 s for Gooog and 3 min 18 s for Luna 3, including tools and waiting, and led to different results. They began at 04:46 and 04:29 respectively. This is not a controlled benchmark of model speed or cost, and the boards do not reproduce hidden reasoning or full tool logs.