Where to find your usage
Click Usage in the left sidebar. You’ll see two numbers at the top: total credits consumed and total requests made, for the selected time range.
Use the 7 Days / 30 Days / 90 Days buttons in the top right to switch between time ranges.
Reading the charts
Credits by Model shows a stacked bar chart of credit consumption per day, broken down by model. Taller bars on a given day mean more activity. The color legend shows which models drove that usage — Claude models, image generation models, and video generation models each appear in different colors.
Requests by Model shows the same breakdown by request count rather than credit cost. A model can appear frequently in Requests but minimally in Credits if it’s cheap to run (like Haiku), or the reverse if it’s expensive (like Opus).
How credits work
Different models and task types consume credits at different rates:
- Opus 4.6 consumes the most credits per request — it’s the most capable model and best for complex reasoning tasks
- Sonnet 4.6 is the default and strikes a balance between capability and cost
- Haiku 4.5 is the cheapest option for quick, simple answers
- MiniMax M2.5 / M2.7 are cost-efficient options for everyday tasks
- Image and video generation (Gemini, Seedream, etc.) consume credits separately from text — check the chart to see how much media generation contributes to your usage
Choosing the right model
The model selector sits in the bottom-right of every chat input. Pick based on what you need:
| Task | Recommended model |
|---|
| Complex analysis, coding, reasoning | Opus 4.6 |
| Most everyday tasks | Sonnet 4.6 |
| Quick questions, simple edits | Haiku 4.5 |
| Cost-sensitive tasks at scale | MiniMax M2.7 |
Max plan
Max plan users have unlimited tokens. The Usage dashboard still shows your consumption for reference, but there’s no cap and no anxiety about running out mid-task.
If you’re watching your credits, run complex multi-step tasks on Sonnet rather than Opus unless you specifically need Opus’s reasoning depth. For research tasks that involve lots of back-and-forth, Haiku handles the lightweight steps well.
