Body
Quick Links: | Article Summary | Understanding ChatGPT Messages and Agentic Tokens | Choosing the Right ChatGPT Model | Choosing a Reasoning Effort | Managing ChatGPT Message Usage | Managing Agentic Token Usage |
Article Summary
Chico State's ChatGPT environment provides several AI models and features designed for different levels of work. Choosing the appropriate model and reasoning effort for a task can help you use your available ChatGPT resources efficiently and avoid reaching usage limits unnecessarily.
This article explains:
- The difference between ChatGPT messages and agentic tokens.
- When to use Instant, Luna, Terra, or Sol.
- How reasoning effort affects resource consumption.
- How to avoid unnecessarily consuming your available ChatGPT messages.
- How ChatGPT Work and Codex consume a separate, shared agentic token allowance.
- How to check your agentic token usage.
- What happens when you reach a usage limit.
The most important guideline is to use the smallest model and lowest reasoning effort appropriate for your task. A larger model or higher reasoning setting is not automatically better for every request. Reserve the most capable models and highest reasoning levels for work that actually requires them.
Understanding ChatGPT Messages and Agentic Tokens
Chico State's ChatGPT environment has two different types of usage that should not be confused: ChatGPT messages and agentic tokens.
ChatGPT messages
ChatGPT messages are used during regular conversations with ChatGPT. Each time you submit a prompt using a metered model or reasoning option, your available ChatGPT message allowance may be affected. More capable models and higher reasoning efforts consume this allowance more quickly.
Chico State users have an allowance of 160 ChatGPT messages within a three-hour window. If you reach the applicable limit, you must wait for capacity to become available again as the three-hour window progresses.
Instant is unlimited and is the best starting point for many everyday ChatGPT tasks.
Agentic tokens
Agentic tokens are separate from ChatGPT messages. They are used by features that perform more extensive or agent-like work on your behalf, including:
ChatGPT Work and Codex draw from the same agentic token allowance. Using one of these tools therefore reduces the amount of agentic capacity available to the other.
The agentic allowance includes both a five-hour limit and a weekly limit. Reaching either limit can temporarily prevent additional agentic work until the applicable usage window resets.
In short:
- Regular ChatGPT conversation = ChatGPT messages
- ChatGPT Work and Codex = agentic tokens
Choosing the Right ChatGPT Model
ChatGPT models are available at different capability levels. Larger models are intended for more difficult tasks, while smaller models are more efficient for routine work. Using a larger model than your task requires can consume your available ChatGPT messages unnecessarily.
For most tasks, start with Instant or Luna. Move to Terra or Sol when the complexity of the task justifies the additional resources.
| Model |
Recommended Use |
Higher Education Examples |
| Instant |
Everyday questions, writing assistance, summarization, brainstorming, and other routine tasks. |
- Draft or improve a message to an instructor, student, colleague, or department.
- Rewrite an announcement to make it clearer.
- Summarize meeting notes.
- Brainstorm questions for a class discussion.
- Improve the grammar or tone of existing text.
- Create a basic outline for a document or presentation.
|
| Luna |
Lightweight reasoning and efficient processing when a task needs more analysis than Instant but is still relatively straightforward. |
- Compare two versions of a course announcement.
- Organize a set of notes into categories.
- Create discussion questions from supplied course material.
- Identify themes in student or departmental feedback.
- Help structure a straightforward academic or administrative process.
|
| Terra |
Moderately complex work requiring a balance of reasoning capability and efficiency. |
- Compare several university policies and identify important differences.
- Analyze a moderately complex dataset and explain patterns.
- Develop a detailed lesson, workshop, or training plan from multiple requirements.
- Synthesize information from several documents.
- Help evaluate several possible approaches to an academic or administrative problem.
|
| Sol |
The largest model, intended for the most complex reasoning, research, analysis, and multi-step problems. |
- Synthesize a large collection of research materials and identify relationships or conflicting findings.
- Analyze a complicated institutional problem involving multiple requirements and constraints.
- Work through a difficult research question requiring substantial reasoning.
- Evaluate a complex technical architecture or troubleshoot a difficult technical problem.
- Analyze a large set of documents where conclusions depend on information distributed across the materials.
|
A useful rule: Start small and move up only when necessary. If Instant or Luna can complete the task successfully, there is generally no reason to use Sol.
Choosing a Reasoning Effort
Model size is only one factor affecting usage. ChatGPT may also allow you to select a reasoning effort. Higher reasoning effort tells the model to spend more computational effort working through your request.
Higher reasoning effort is useful when a problem genuinely requires careful multi-step analysis, but it is not automatically better for ordinary work.
- Start with Instant for routine work. Drafting, rewriting, summarizing, brainstorming, and straightforward questions generally do not require a reasoning model.
- Use Luna with a light reasoning effort when some analysis is needed. This can substantially reduce resource consumption compared with using a larger model.
- Move to Terra when the task becomes moderately complex. Use it when Luna is not providing enough reasoning depth.
- Use Sol for genuinely difficult tasks. Sol is intended for complex research, analysis, technical problems, and other work requiring the strongest model.
- Use Ultra effort selectively. Sol combined with Ultra reasoning can consume your available messages very quickly. Reserve this combination for problems where maximum reasoning is actually necessary.
For example, asking ChatGPT to improve the wording of a message to an instructor does not require Sol at Ultra effort. A task involving the synthesis of numerous research sources, competing evidence, and a complex set of requirements may justify a larger model and higher reasoning effort.
Managing ChatGPT Message Usage
Chico State users receive 160 ChatGPT messages per three-hour window for metered ChatGPT usage. Your model and reasoning choices affect how quickly you consume this capacity.
To make your available messages last longer:
- Use Instant whenever it meets your needs. Instant is unlimited and is appropriate for a large percentage of everyday university work.
- Use Luna for light reasoning. Do not automatically select the largest model because it is available.
- Use Terra for moderately complex analysis.
- Reserve Sol for difficult work.
- Avoid using Ultra reasoning for routine tasks. Higher reasoning effort can consume available capacity significantly faster.
- Increase model size or reasoning effort only when the quality of the smaller option is insufficient for the task.
If you reach a ChatGPT usage limit, you will need to wait for availability to return within the applicable three-hour window. Chico State ChatGPT administrators cannot increase an individual user's message allowance, reset the usage window, or provide additional messages.
These resource allocations are part of the ChatGPT service provided to Chico State through the CSU system. Users receive the same available resource allocation under the service. Local ChatGPT administrators do not have controls that can override these usage limits.
Managing Agentic Token Usage
Agentic tokens are consumed separately from regular ChatGPT messages. They are used for agentic features such as ChatGPT Work and Codex.
Because ChatGPT Work and Codex share the same agentic token allowance, significant use of one can affect how much capacity remains for the other.
Agentic usage is governed by both a five-hour usage limit and a weekly usage limit. If you reach the five-hour limit, you must wait for capacity to become available again within that window. If you reach the weekly limit, additional capacity will not become available until the applicable weekly usage period resets.
Check agentic token usage in Codex
- Open Codex.
- Type /status.
- Review the displayed usage information to see your current consumption.
Check agentic token usage on the ChatGPT website
- Open ChatGPT.
- Select your account or profile controls in the lower-left corner.
- Select Settings.
- Select Usage.
- Review your current agentic usage.
You can also view agentic token usage directly at ChatGPT Codex Usage Settings.
Chico State ChatGPT administrators cannot add agentic tokens, reset a user's consumption, extend the five-hour or weekly limits, or override a usage cap. If you reach an agentic usage limit, you must wait for the applicable usage window to reset.
Still need help? Add your department's information or a link to a ticket type for further troubleshooting.
Please do not leave the comment section blank! Provide constructive feedback to make this page better.
```