Table of Contents
Reasoning Mode gives you control over how much effort the AI puts into thinking through each response in ONEchat. Choose from three levels of reasoning depth to match the task at hand, whether that's a quick lookup that needs a fast reply or a complex problem that benefits from careful, step-by-step analysis.
Overview
When chatting in ONEchat with a model that supports reasoning, a Reasoning option appears in the plus menu (+). This control lets you choose how much effort the AI puts into thinking through its response before answering.
Reasoning Mode has three settings:
|
Mode |
Best For |
Speed |
Token Usage |
|
Fast |
Simple questions, casual chat, quick lookups |
Fastest |
Varies* |
|
Balanced (default) |
Everyday tasks that need some depth without a long wait |
Moderate |
Varies* |
|
Thinking |
Complex problems, detailed analysis, multi-step reasoning |
Slowest |
Varies* |
*Token usage is not directly proportional to reasoning level and can vary considerably depending on the prompt and how the model chooses to respond. A faster, more direct response can sometimes use more tokens than a slower, reasoning-heavy one.
Balanced is selected by default for all models that support Reasoning Mode. You can change the mode at any time during a conversation, and it will apply to your next message.
How to Use Reasoning Mode
- Open a conversation in ONEchat.
- Click the plus menu (+) in the chat input area.
- Look for the Reasoning row.
- If the model supports reasoning, you'll see the three mode options: Fast, Balanced, and Thinking.
- If the model does not support reasoning, this option will appear greyed out.
- Select your preferred mode.
- Type your message and send. The AI will respond using the reasoning level you selected.
You can switch modes between messages — there's no need to start a new conversation.
Reasoning Mode Compatibility
The table below lists all available models and whether they support Reasoning Mode. For unsupported models, the Reasoning control appears greyed out in the plus menu. Balanced is the default mode for all supported models.
|
Model |
Reasoning Mode Supported |
|
gpt-5+ |
Yes |
|
claude-sonnet-4-x+ |
Yes |
|
claude-opus-4-x+ |
Yes |
|
claude-haiku-3-5+ |
Yes |
NOTE: Model availability varies by region and instance configuration. If a model listed above does not appear in your instance, it may not be deployed in your region.
When to Use Each Mode
Fast
- Quick factual questions ("What's the capital of France?")
- Simple reformatting or rephrasing tasks
- Casual conversation where speed matters most
- Situations where you want a quicker reply
Balanced (Default)
- General-purpose tasks like drafting emails, summarizing documents, or answering moderately complex questions
- Day-to-day use where you want a good balance of quality and response time
- Most users will find this mode works well for the majority of their interactions
Thinking
- Complex analytical problems that require step-by-step reasoning
- Multi-part questions where accuracy is critical
- Tasks involving math, logic, code debugging, or detailed comparisons
- Situations where you'd rather wait longer for a more thorough, accurate answer
Why Token Usage May Vary Across Reasoning Settings
While reasoning settings influence how the model approaches a problem, they do not directly control the number of tokens used. Token usage can vary based on how the model chooses to generate a response.
In some cases, a faster or more direct response may involve pulling in a larger amount of external or reference content, which can increase token usage. In other cases, a more deliberate or "reasoning-heavy" response may rely more on internal processing and generate fewer tokens overall.
As a result, reasoning level and token consumption are not directly proportional, and it is expected to see variation depending on the specific prompt and how the model decides to respond. In practice, this means a lower reasoning setting such as Fast can sometimes use the same or more tokens than Thinking — particularly when the model reaches out to tools or external content to answer quickly.
The effect of the reasoning setting also depends on the prompt:
- Simple, objective prompts tend to show little difference in total token usage between reasoning levels.
- More complex prompts and workflows show a more pronounced difference between levels.
A helpful way to think about it: Picture reasoning as time spent, not tokens spent. The quickest answer might be "let me ingest this whole Wikipedia page," which is thousands of tokens — versus "let me ruminate over this for 30 seconds without reaching out to a web page at all," which can be far fewer. The mode shapes the model's approach, and the token count follows from whatever that approach requires.
Good to Know
- Your selection persists within a conversation. Once you choose a mode, it stays active until you change it. Starting a new conversation resets to Balanced.
- Reasoning Mode does not predictably raise or lower token usage. A higher reasoning level does not reliably consume more tokens, and a lower level does not reliably consume fewer. Token usage depends on the prompt and how the model chooses to respond.
- System agents use admin-configured defaults. If you're chatting with a system agent (one configured by your organization's admin), the reasoning level is set by the admin and cannot be changed by the user via the plus menu.
Frequently Asked Questions
Q: I don't see the Reasoning option in my plus menu. Why?
The Reasoning option only appears when your current model supports reasoning. Try switching to a supported model (e.g., GPT-5+, or Claude Sonnet 4+.
Q: Does a higher reasoning level always use more tokens?
No. Reasoning level and token consumption are not directly proportional. In some cases a lower setting like Fast can use the same or even more tokens than Thinking — for example, when the model answers quickly by pulling in external or reference content. Token usage depends on the specific prompt and how the model decides to respond, so variation is expected.
Q: Does Thinking mode always give better answers?
Not necessarily. For simple questions, Thinking mode may produce a longer response without meaningfully improving accuracy. It's most valuable for complex, multi-step problems. For straightforward tasks, Balanced or Fast will give you a great answer more quickly.
Q: Does changing the reasoning mode affect my conversation history?
No. Changing the mode only affects future messages. Previous responses in the conversation are not regenerated.