# From tokens to output cost

> Two side-by-side visual treatments explain tokenization, attention, next-token selection, and output cost using word chips, connection diagrams, probability bars, counters, and short-versus-long answer comparisons.

- Kind: Video
- Page: https://meetreference.com/v/prompt-cost-explainer-two-styles/
- Ready-to-use agent handoff: https://meetreference.com/v/prompt-cost-explainer-two-styles/agent.txt
- Exact prompt as plain text: https://meetreference.com/v/prompt-cost-explainer-two-styles/prompt.txt
- Authorship: AI-made, per the source
- Made with: Not stated
- Model: Claude Opus 5.5, GPT-6 Sol
- Built with: HyperFrames
- Style: Explainers; technical explainer, dark mode, token diagrams, modular cards, data visualization, split screen
- Length and format: 20 sec, 1280:440 (1280x440 source)
- Playback: a muted visual reference. The original soundtrack is available only at the source.
- Shared by @TheHunterBohm in X, Sep 23, 2026: https://x.com/TheHunterBohm/status/2102724864205566388
- Prompt fidelity: Verbatim. This is the complete prompt exactly as the creator published it.

## Before you start

Guidance from Reference, separate from the creator's instructions.

1. Read the reference and inspect the project, installed libraries, and available agent skills before writing code. Reuse suitable tools already present.
2. Prefer the creator-reported libraries and skills when they fit the task. Explain what useful additions help with, and install missing project dependencies or relevant skills from official sources when the environment and the user's existing permissions allow it. Respect the project's package manager and setup rules; do not ask again for routine setup already authorized.
3. Distinguish tools used by the creator from your own recommendations. A listed tool is not automatically required, and an empty tools or requirements list means the setup is unknown, not that nothing is needed.
4. Ask concise questions when missing inputs or choices materially affect the result: paid services, credentials, the user's assets, or a different production approach. Use existing answers and permissions; continue independent work while waiting.
5. If a useful library, skill, or service cannot be installed or used, explain an available alternative and its expected effect on quality, features, or effort. Get the user's choice before a material compromise; otherwise state the fallback and continue. Never silently simplify the result or claim to have used a tool you did not use.

## Recreating this video

These are the creator's published instructions, unedited.

## Prompt

```text
Make a 20-second motion graphic that explains how an AI language model turns a prompt into an answer, and why a long answer costs more… The charts then break apart into pixel blocks, and those pixel blocks rebuild into the explanation.
```

## What it looks like

Written by Reference from the video frames, not by the creator.

A paired explainer compares Opus 5.5 and GPT-6 Sol using token-count examples, charts, and cost summaries.

- Color: Opus 5.5 uses near-black and deep navy with restrained teal, yellow, and purple accents. GPT-6 Sol uses pale gray and white panels with navy backgrounds, blue interfaces, and teal, green, and yellow highlights.
- Type: Clean sans-serif interface typography, with bold titles, small labels, and prominent numeric callouts.
- Framing: Each frame is divided into labeled left and right panels. The left side is sparse and dark; the right side presents denser dashboard-style cards and diagrams.
- Structure: Usage metrics lead to tokenized sentences, attention diagrams, candidate probabilities and output counters. Short and long answer examples conclude both visual treatments.

Storyboard, nine frames spread across the video: https://meetreference.com/media/prompt-cost-explainer-two-styles/storyboard.webp

- Frame 1 at 1.1s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f1.webp
- Frame 2 at 3.3s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f2.webp
- Frame 3 at 5.6s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f3.webp
- Frame 4 at 7.8s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f4.webp
- Frame 5 at 10.0s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f5.webp
- Frame 6 at 12.2s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f6.webp
- Frame 7 at 14.4s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f7.webp
- Frame 8 at 16.7s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f8.webp
- Frame 9 at 18.9s: https://meetreference.com/media/prompt-cost-explainer-two-styles/f9.webp

## How it was made, per the creator

- Notes: Creator compares Opus 5.5 and GPT-6 Sol at medium effort. Reported Opus run: 12.7 minutes, $2.86; Sol run: 11.4 minutes, $1.09.

## Links

- Watch: https://x.com/TheHunterBohm/status/2102724864205566388
- Original post: https://x.com/TheHunterBohm/status/2102724864205566388
- Original prompt: https://x.com/TheHunterBohm/status/2102724876092277172