What Are Tokens?

A practical example

Say you are working on a 2,000-word report. That is roughly 2,700 tokens. Your Builder reads the full document plus the prompt (~500 tokens) = about 3,200 tokens in, then writes back the updated version at roughly the same length. That is ~6,000 tokens per round.

Over 5 rounds, that is ~30,000 tokens from your Builder alone. With a paid API key this costs a few cents. On a free tier with strict limits, this could hit the cap by round 2 or 3.

Bottom line: use a paid API key for your Builder, or use Gemini's free tier which is genuinely generous while your Google AI Studio account has billing disabled. Keep cost-conscious or free-tier AIs as Worker Bees where the token load is light.

Gemini's free tier vs. paid tier — read this if you've added a credit card

Gemini's free tier is genuinely generous, but only while your Google AI Studio account has billing disabled. The moment you add a credit card and enable billing — even if you intended to keep using Gemini for free — requests can route through paid-tier paths and charge per token. The Gemini model string in WaxFrame (gemini-2.5-flash) doesn't change; what changes is the billing path Google's API selects for your project.

This matters most for the Builder role. Builders read project setup, reference material, the full working document, and every reviewer's suggestions every round, then write the document back out. That's the highest-token role in WaxFrame by a wide margin — a multi-round session with paid-tier Gemini as Builder on a long document with reference material attached can chew through a few dollars far faster than you'd expect.

If you've enabled billing on AI Studio: consider keeping Gemini as a Reviewer (lighter token load) and using DeepSeek or another low-cost paid model as your Builder. Check the AI Studio usage page after longer runs to confirm where your spend is going.

Tips for managing token costs

Keep your document focused. The longer your document, the more tokens your Builder uses every round. Trim content that is no longer relevant as the document improves.

Switch Builders if a round fails. DeepSeek is the best value for most documents. If a round fails with a missing output structure error — especially on a large or complex document — try switching to ChatGPT or Gemini as your Builder and retrying. Some AIs are less consistent at following strict formatting instructions under heavy load. This is a behavioral difference between models, not a hard size limit.

Use Worker Bees wisely. You can disable individual AIs per session from the work screen. If an AI is consistently unhelpful, turn it off — fewer Worker Bees means less input for the Builder to process.

Start small. For your first session, try a shorter document or a simpler goal. Once you understand how the rounds work you can take on larger projects confidently.

Reference material and your token budget

If you have used Setup 4 — Reference Material to paste an RFP, job description, style guide, or scoring rubric, that material is sent to every reviewer every round alongside your document. The hive needs to see it on every pass to cite against it, and that has a real cost.

The math. A 5,000-character RFP is roughly 1,250 tokens. A six-bee hive running four rounds spends about 30,000 tokens just transmitting the reference material — before any document content, project goal, or AI response is counted. That is on top of the per-round Builder cost above.

Two practical consequences.

1. Trim aggressively. Paste only what is most important. If a 40-page RFP has three pages of substantive requirements and 37 pages of boilerplate, paste the three. The hive cannot read what you do not paste, but it also costs you money for every token you send.

2. Pick a cost-conscious Builder when reference material is large. DeepSeek and Gemini Flash are the cheapest options for high input-token loads. The Builder receives the largest prompt every round, so this single change has the biggest cost impact on a reference-material-heavy session. If you are running an RFP response with 10,000+ characters of reference material, a low-cost Builder is almost always the right call.

The chars / words / tokens counter on Setup 4 and on the Work-screen Reference drawer gives you a live readout. Use it to gauge order of magnitude before committing to a Builder.