AI models, settings, and credit usage
Written By Johan Hanegraaf
Last updated About 22 hours ago
Choose AI settings that suit your design task, and use examples from Arkio projects to understand how credit usage can grow from a small change to a developed model.
A good starting point: Auto, Medium reasoning, and Normal speed. Review the first result before asking for more detail or presentation images.
What affects AI credit usage?
Credits depend on the work the AI performs. One request can involve inspecting the scene, creating geometry, checking the result, and making corrections. A short prompt can therefore use more credits than its length suggests.
Scope and detail: A material change, a furnished room, and a complete building require different amounts of work.
Model, reasoning, and speed: These settings affect the processing used for your request.
References and conversation history: Understanding drawings, images, and an evolving design adds work.
Iterations: New alternatives, refinements, and corrective follow-ups add to the project total.
Presentation: Generating images, videos, and using voice adds usage beyond modeling.
For balances, allowances, and purchases, see AI credits and your work.
How to change your settings
Use the controls inside the chat message box:
Select Model to choose Auto, Sol, or Astra.
Select Reasoning to control how much effort the model spends working through your request.
Select Speed to choose Normal or Fast. Selecting Astra also makes Ultrafast available. Normal uses fewer credits; faster options increase credit usage.
On smaller screens, these controls may appear as icons. Changes apply to your next request; a task already running keeps the settings it started with.
Which model should I use?
For example, choose Astra to develop a furnished interior from a floor plan, or Sol to change materials and save a viewpoint. Choose Auto if you would prefer Arkio to make that choice.
Let Arkio choose with Auto
Auto selects between Sol and Astra when each task starts, aiming to use the lower-cost model that can handle your request well. Routine changes and scene inspection generally use Sol, while demanding design work generally uses Astra.
The chat displays Auto: Sol or Auto: Astra so you can see which model is working. That model stays with the task until it finishes. Auto keeps your selected reasoning level and speed.
Choosing a reasoning level
Available levels are Low, Medium, High, XHigh (Extra High) and Max. Higher reasoning levels generally consume more AI credits and can increase completion time.
With Astra, Low or Medium reasoning is already capable of handling most design tasks. Use High, XHigh, or Max for your most demanding requests, such as interpreting complex references or coordinating many design constraints.
Start with a lower reasoning level and increase it when the task needs more careful problem-solving. There is no fixed credit multiplier between Low, Medium, and High: the difference depends on the task and the work needed to complete it.
Choosing a speed
These comparisons are for comparable AI processing with the same model. Total credit usage and completion time depend on the task, reasoning level, and any additional generation features used.
Ultrafast requires selecting Astra directly. Auto and Sol offer Normal and Fast.
Speed and reasoning serve different purposes: reasoning controls the effort spent on your request; speed controls how quickly that work is processed. Choosing Fast or Ultrafast does not replace choosing a more capable model or an appropriate reasoning level. Faster AI processing does not guarantee that the entire design task finishes by the same multiplier.
Examples of everyday tasks
The examples below come from our sample-project sessions. The modeling amounts below are approximate estimates, not verified charges or fixed prices.
Each row is one observed request. These examples do not establish an expected range for every similar edit, and their model and reasoning settings are not consistent. Use them to understand the scale of individual changes, rather than to compare Sol and Astra.
From a first concept to a developed project
The first concept is only one stage of a design session. Developing interiors, refining surroundings, exploring alternatives, and adjusting details all contribute to the total credit usage.
The table below shows estimates from example projects recorded in October 2026. Use these as a guide, rather than a fixed budget. Actual usage will vary depending on the complexity of your request, your AI model and settings, and the changes needed to achieve your desired result.
The completed project estimate includes the first concept and follow-up work; do not add the two columns together. Separate image and video generation charges are excluded. These are approximate guides to the example workflows.
Kitchen renovation: develop a room from references
Starting with a floor plan and room photos, the request created an open-plan kitchen and dining concept with a central island, integrated cabinetry, and warm materials.
First concept: ~750. Completed: ~1,000. Follow-ups added the missing sink, small appliances and details. This is a useful example of a defined room-sized exercise with a small number of refinements.
Summer Cabin: develop a building and its setting
The initial brief described a 90 m² family cabin with two bedrooms, panoramic glazing, sheltered terraces, and an open living area. The sample also explores a pond, pier, access road, parking and a wooded landscape.
First concept: ~1,000. Completed: ~3,000. This illustrates how development beyond the initial building can become a substantial part of the exercise.
Townhouses: interpret a drawing and create variations
The first request used a façade drawing to create a townhouse and four matching neighbors. Later requests introduced façade and color variations, basic interior walls, and garden and street details.
First concept: ~1,250. Completed: ~2,500. Reusing a façade provides a starting point, while varying the buildings and developing interiors adds work.
Residential Neighborhood: explore a larger concept plan
The initial request created a neighborhood of 26 buildings containing 34 homes with streets, paths, planting and varied housing volumes. Follow-ups refined connections and viewpoints and requested presentation outputs.
First concept: ~1,500. Completed: ~2,000. The editable scene remains a conceptual plan without much detail. American-style houses were requested in the renders. Its lower estimated credit usage shows why site area alone is not a reliable guide to credit usage.
Six-Storey Apartment: develop massing into a detailed building
The initial request created a six-story courtyard building with rounded balconies, recessed terraces, and a residential street setting. Follow-ups developed stairs, landings, and courtyard-facing corridors, then added apartment layouts, kitchens, furniture, and shared spaces.
First concept: ~1,500. Completed: ~6,000. Later refinements aligned balcony dividers with apartment boundaries, added terrace doors, adjusted finishes and introduced planting. This example shows how circulation, repeated interiors, and detailed corrections can account for much of a building's development cost.
Open these projects from Samples in Arkio Web to inspect the results. Their example conversations are shortened explanations of the workflow, not complete records of every original request.
Budget separately for images, videos, and voice
Keep modeling and presentation as separate parts of your budget. Use the estimates below as a starting point when planning images and video for your design.
These are approximate planning estimates, not fixed prices. Actual usage depends on the chosen model, image size and quality, video resolution and duration, and any additional generation attempts. Live rendering and voice have their own usage and are separated from any AI tasks requested.
Example presentation budget: Four AI renders would use approximately 20–60 credits. Adding one 10-second video brings the estimated media total to about 120–160 credits, before any separate agent processing or additional attempts. These media estimates are additional to the modeling figures above.
Plan your work and review your usage
Define the first result. Specify the scale, key spaces, references, and level of detail you need. For example: “Create the cabin layout and main openings; we will add landscape details after reviewing it.”
Choose suitable settings. Start with Auto, Medium reasoning, and Normal speed. Use the model guidance above for more demanding design work or clearly defined edits.
Review before adding detail. Check the layout, dimensions, and main geometry before developing furniture, planting, and small objects.
Make focused follow-ups. Select the objects involved and explain what should change and what should stay.
Plan presentation outputs. Decide which views you need before requesting renders or video.
Check your balance as you work. Open Account → AI credits. Temporary reservations are not final charges. Stopping a request does not reverse work already performed
Was this helpful?
Still need help? Ask the team