With AI Studio becoming available for everyone, a lot changes. Not all of it obvious, and with important implications.
Most importantly, if you don’t pay attention you might run out of credits, which can cause all of your AI Studio rules to stop working.
This article aims to inform you of the current workings of AI Studio credit consumption, and give you tips on being credit-efficient to get the most bang for buck.
Index:
1. How it works
2. The specifics of credit consumption
3. How to use AI Studio effectively and efficiently
What’s Next?
1. How it works
First, let’s lay out the definitions and rules of the game.
Disclaimer: Although I think this article accurately captures the current state, this functionality is certainly evolving quickly. For most up to date info visit Asana Help Center - AI Studio Add-on and Pricing
If you do spot something that is incorrect or out of date I’d appreciate a comment. This helps me keep the article up to date.
1.1. Turning AI Studio on or off
Admins can turn AI Studio on or off in the Amdin console:
Settings > Asana AI > Opreation settings > AI Automations
1.2. Credits
The relatively high energy/GPU-time consumption of AI makes it more expensive than normal rules. Asana uses credits to indicate how much of these resources your AI rules consume.
The amount of credits consumed in the current period can be checked in the Admin console under Billing > AI Studio
You can track the amount of credits a specific rule consumes using Rule execution history
1.3. AI Studio Pricing options
Here are the variants of AI Studio that Asana currently offers:
| Option | Plan | Free | Credit limits | Resets | Build list | Credits per year | How to purchase: |
|---|---|---|---|---|---|---|---|
| Basic | Starter |  yes |
50k | monthly |  no |
0.6M | n/a |
| Basic | Advanced | yes |
75k | monthly | no |
0.9M | n/a |
| Basic | Enterprise | yes |
200k | monthly | no |
2.4M | n/a |
| Plus | Starter and above | no |
Starts at 100k | monthly | no |
Starts at 1.2M | Admin console |
| Pro | Starter and above | no |
Starts at 5,000k | quarterly | yes |
Starts at 20M | Contact Sales |
The amount of credits you get to spend are the credits included in basic AND the amount of credits purchased with the paid add-on.
For Plus and Pro additional packages can be purchased. Additional packages have the same amount of credits as the initial package for the same plan, but are significantly cheaper.
The Pro plan is the only plan that allows limiting the consumption of the purchased credits to a builder list. The rest (all users not on the build list) will consume the credits included with basic.
1.4. What happens when you run out of credits?
Assume the following:
- At 80% of credit consumption:
– admins will get an email warning - At 100% of credit consumption
– admins will get another email
– rules using AI Studio will no longer run*
– users that have AI studio rules that no longer work will get an in-app notification
*A notable exception, as communicated to me by Asana:
For Enterprise(+) plans with AI Studio basic: we automatically increase the credit limit the first time a customer hits it, and notify them by email. This one-time increase is intentionally generous — up to 3x the original allocation — to give teams time to adapt, either by purchasing AI Studio credits or adjusting their workflows. Starting the following month, credits will be shut off as soon as 100% of the limit is reached.
Related product feedback: Add AI Studio credit limit notification at 50%
1.5. How to make AI-Studio rules start working again
After you’ve run out of credits, there are three things that make AI-Studio rules start working again:
- Purchase additional credits
- Move a users’ consumed credits into the Pro pool by adding them to the builder list - assuming that pool hasn’t also run out of credits.
- The current credits expire, and a new batch arrives - monthly or quarterly, depending on the plan
2. The specifics of credit consumption
Now that we know the framework, let’s dive into the specifics that determine credit consumption.
There are three main factors determining credit consumption: Input, Output, and AI Model.
In short,
- Input is what is given to the AI model
- Output is what it gives back
- AI Model is the type of AI utilised.
2.1. Input
The amount of information AI is given, including:
- AI’s guidance/instructions, including linked or uploaded documents
- All attributes of the task (See Asana AI Studio FAQ - Data and security for extended list)
- If enabled in rule settings:
– Activity
– Information from linked tasks
– Information from projects@mentionedin instruction (extent depending on rule’s AI Studio settings) - If web access is enabled:
– Information behind URLs in Task
– Information behind URLs in Guidance
2.2 Output
Ouput is what AI gives back, including:
- Evaluating conditions
- The actions it takes, including the text it generates.
- If web-access is enabled: The work it does searching the web
2.3 AI Model
The type/version of AI that you choose to handle your AI Studio rules is called the AI Model. Simpler models consume less energy/GPU-time, and therefore less credits, and more complex models consume more credits. More complex models are also capable of handling more input data, for instance larger documents.
| Model | Input multiplier | Output multiplier | Understands images |
|---|---|---|---|
| Claude Sonnet* 4 | 5 | 25 | yes |
| Claude 3.5 Sonnet | 5 | 25 | yes |
| GPT-5 | 2.5 | 20 | yes |
| GPT-4o | 5 | 25 | no |
| GPT-5 mini | 0.5 | 5 | yes |
| Claude Opus 4 | 25 | 125 | yes |
| o1 | 30 | 150 | no |
| o3-mini | 6 | 30 | no |
| Claude 3.5 Haiku | 3 | 15 | yes |
| Claude 3.0 Haiku** | 1 | 5 | yes |
| GPT-4o mini | 1 | 5 | no |
* current default
** The use of Claude 3.0 Haiku in Smart workflows has been deprecated. Existing rules utilizing this model will still run, but Asana plans to migrate these rules to Claude 3.5 Haiku in the future.
In AI Studio rules, you can choose the model AI Studio uses in the rule guidance.
Source for model specific multipliers: AI Studio add-on and pricing - model pricing
2.4 Estimating credit consumption.
There is almost always variation in credit consumption, even if the same rule is fired twice with in the same way. Here we focus on the aspects that can be controlled.
The following formula illustrates the relationship between the different elements that determine credit consumption:
(input * input_multiplier) + (output * output_multiplier)
Note that:
- The difference between the cheapest and expensive models is a factor of 30(!).
- Output generally costs 5 times as much as input.
2.5 Testing results
If you’re interested, here are the results of my testing comparing the credit consumption of the different models.
🧪 Testing results
This was my testing setup:
- Rule: When a new task is added to this project, add comment in Task (
Use AI) - Guidance: “Give a helpful comment”
- Task title: Explain Asana
I only varied these factors:
- AI Model
- Web access on/off
Here are the results I found:
2.5.1 Average credit consumption
| Model (input:output) | no web | with web |
|---|---|---|
| 4o (1:5) | 36 | 537 |
| Claude 3.5 Haiku (3:15) | 118 | 786 |
| Claude Sonnet 4 (5:25) | 270 | 1388 |
I’m guessing the impact of web access is relatively big in this scenario, as I’ve tested with very little input data.
2.5.2 Standard deviation in credit consumption:
| Model (input:output) | no web | with web |
|---|---|---|
| 4o (1:5) | 9 | 289 |
| Claude 3.5 Haiku (3:15) | 1 | 36 |
| Claude Sonnet 4 (5:25) | 68 | 136 |
2.5.3 Raw data
| test nr | 4o | 4o +web | Claude 3.5 Haiku | Claude 3.5 Haiku +web | Claude Sonnet 4 | Claude Sonnet 4 +web |
|---|---|---|---|---|---|---|
| 1 | 31 | 410 | 119 | 767 | 246 | 1213 |
| 2 | 54 | 485 | 117 | 786 | 246 | 1339 |
| 3 | 31 | 411 | 120 | 778 | 246 | 1395 |
| 4 | 31 | 391 | 118 | 754 | 245 | 1218 |
| 5 | 31 | 391 | 117 | 737 | 247 | 1323 |
| 6 | 31 | 413 | 120 | 752 | 247 | 1425 |
| 7 | 32 | 417 | 119 | 780 | 250 | 1415 |
| 8 | 52 | 394 | 118 | 818 | 475 | 1530 |
| 9 | 31 | 698 | 118 | 839 | 250 | 1694 |
| 10 | 32 | 1364 | 118 | 851 | 246 | 1324 |
3. How to use AI Studio effectively and efficiently
Now that you know the rules of the game, we’ll dive into how to play it well.
3.1. Defining effectiveness and efficiency.
Let’s start with some definitions, as many people don’t understand what these terms really mean.
Effectiveness is your ability to achieve desired results.
This is why effectiveness should always be considered first. Because being more efficient is useless if the result is unwanted. That’s like caring about whether your cab takes the shortest route when he’s driving you to the wrong destination.
For further insights into effective use, I’d advise to take the AI Studio Foundations Skill badge if you haven’t already.
Efficiency is: results_achieved / resources_consumed
Resources - in this instance - are credits. Improve this ratio, and you are more efficient.
And I’ll give you a mantra to keep in mind:
“Make it work, make it right, make it fast”
– Kent Beck
- Make it work: Get the result
- Make it right: Make sure your colleagues can logically understand how this works (or yourself in a couple of weeks for that matter)
- Make it fast: Make it efficient.
3.2. How to be effective and efficient.
Now let’s dive into the levers you can pull to be more effective and efficient.
3.2.1. Don’t use AI when you don’t need to.
AI can do a lot of things, but it’s by no means a silver bullet. So be very selective in applying AI. This starts with the decision to use AI Studio in a rule or not, and also for deciding whether to utilise AI in individual rule-elements.
Also note that Asana has Script actions. These allow you to do things normal rules can’t do, but are straightforward enough not to need AI to tackle them. To learn more about how this may complement your rules check out @Arthur_BEGOU ’s article: Use Script Actions to Make the Most of your AI Studio credits
3.2.2. Choose the right model and data permissions.
The biggest impact on efficiency - by far - is configuring an AI Studio rule’s settings, which can be accessed at the button with sliders, next to the Guidance button.
The credit consumption between models can differ with a factor 30. Choose the simplest model you need (see: 2.3. AI Model). Make sure to confirm correct working with testing.
Only allow the Data and permissions that it needs.
3.2.3. Don’t enable web-access when it’s not needed.
Web access takes a lot of credits, as it impacts both input and output. It can be really valuable when you use it for research, but don’t rely on it when you don’t need to. The AI’s have a lot of training data, including knowledge of Asana, for which web access is not required.
If there is public information you would need regularly, it might be worth storing or referencing in the prompt instead of searching for it. For example, someone mentioned a workflow that used the australian fiscal year. I’d go about it like this:
- Test whether the model knows this without needing web access.
- If not, include it in your prompt: “The Australian fiscal year runs from July 1 to June 30 of the following year.”
Either cases, you won’t need web access anymore. (Assuming the model is smart enough to interpret this)
3.2.6. Mind the frequency
One last obvious thing to consider is how frequent you expect the rule to run. The more frequent the rule execution, the more important the efficiency.
What’s next?
Behind the scenes I have an active conversation with Asana about improving how this works.
Yesterday (at the London Work Innovation Summit) I even had the opportunity to sit down with Asana’s AI Engineering lead. She informed me of many planned changes aimed at improving our insight into credit consumption, and our control over it.
Although I cannot share any specifics or timelines I am excited about the direction, and I walked away confident that Asana is working hard to make managing this easier.
I’ll do my best to keep this topic up to date with all relevant information and developments.
Jan-Rienk Hemminga - Asana expert and partner for Improving Every Day