Cost and limits
Who bills you, where to watch the running total, and what drives it up.
Excelente is free and has no in-app purchases. What you spend is what the model costs, billed by OpenRouter directly.
Who charges what#
| Included model | Your own key | |
|---|---|---|
| Who pays | A.CRE, or CRE Agents, Inc. | You |
| Billed by | Nobody | OpenRouter, at its published rates |
| Limits | Free, limited usage each month, set by A.CRE | Whatever you set at OpenRouter |
CRE Edge does not resell model access and does not process payments. The relationship is between you and OpenRouter.
Watching the total#
The header carries an info popover showing your running session cost. It also lists the active model, the reasoning level when it is not Off, prompt cache statistics, the build version, and a Reset session cost button.
/cost prints the same figure into the transcript:
💰 Session cost — No usage recorded yet this session.
💰 Session cost — $0.42m (sub-millidollar)
💰 Session cost — $0.1234
Sub-millidollar totals are shown in millidollars rather than rounded to $0.00, because "$0.00" on a
run that cost something tells you nothing.
On the included model the cost row carries an asterisk:
* Covered by your membership's included usage.
What drives cost#
In rough order of impact:
- Model choice. Frontier models cost multiples of mid-tier ones. The Best value sort in the model explorer exists for this.
- Reasoning level. Reasoning tokens bill as output. High on a long build adds up.
- Conversation length. Every turn resends the conversation. A long session costs more per turn than a fresh one, which is a good reason to start a New chat when you change subject.
- Screenshots and PDFs. Images are expensive. A 90-page PDF attached whole is a large bill for the four pages you needed.
- Sub-agents. Each one is its own conversation. Worth it for verification, and not free.
Prompt caching#
Where the provider supports it, the stable head of your conversation (system prompt, skills, conventions) is cached, so repeat turns bill a fraction for those tokens. The header popover shows cache statistics so you can see it working.
Caching is the reason a long conversation in one workbook is cheaper per turn than the token counts suggest. It is also why switching models mid-conversation costs a full re-read on the next turn.
Controlling spend#
- Set a limit at OpenRouter. Their dashboard has spend caps. Use one.
- Match reasoning to the work. Off for reading, High for building.
- Start a new chat when the subject changes. History keeps the old one.
- Attach the pages you need, not the whole document.
- Turn off connectors you are not using. Loaded tools are schema on every turn.
- Enable the skills the job needs, not all of them.
- Point the Summary model at something cheap. Compaction does not need frontier reasoning.
Included usage#
For a limited time, CRE Agents members and A.CRE Intelligence Hub members get free, limited usage of the included model each month. Each connected membership adds to the limit. It resets on the first of every month, whatever day you joined, and does not carry over. The ring in the composer, left of the connector marks, fills as the month's usage is used; click it for the percentage left and the refresh date. The same figures are in Settings under Memberships.
When the month's included usage is used up the included model stops answering until the month resets, and the message says so. The included model and its limit are offered at A.CRE's discretion and may change or be withdrawn at any time. Add your own OpenRouter key to keep working, with better models, control over reasoning, and independent Reviewer verification on top.
Nothing runs unattended#
Excelente only calls a model in response to something you did. It does not poll, does not run in the background, and does not spend anything while the pane sits open. Closing the pane mid-run stops the run.