ADITO-LLM Now Supports Thinking
The ADITO-LLM has a new capability: thinking mode.
It is off by default. You turn it on with one parameter. This post explains what it does and when to use it.
What thinking mode does
When thinking mode is enabled, the model works through the problem before responding. It reasons step by step instead of answering immediately.
For simple tasks this makes no difference. For complex ones it does: ambiguous inputs, multi-step analysis across records, anything where getting it wrong costs time to fix.
How to enable it
Add enable_thinking: true to your request body:
{
"model": "adito-llm",
"enable_thinking": true,
"messages": [...]
}
That is it. The default is false, so existing integrations are unaffected.
One caveat: thinking mode is slower and uses more tokens. Only enable it where the quality gain is worth it.
When to use it
Worth enabling:
- Classification with overlapping or ambiguous categories
- Analysing long, unstructured text
- Multi-step reasoning across several CRM records
- Anything where a wrong answer is expensive to correct
Not worth it:
- Simple completions
- Short extraction tasks
- Anywhere response time matters more than depth
Where to go next
Full parameter reference and code examples are in the ADITO-LLM API reference.
./Flo
