Flick through the images and you will see that the Anthropic chat node has a setting where you can 'Enable Thinking'
When you turn this on your agent actually spends tokens 'thinking' about your question, just like what you get in the Claude chat interface.
The last image shows how the output response contains two objects:
• One is the type: thinking - the thinking tokens • The other is type: text - the actual response
The OpenAI chat node also has a similar setting for 'Reasoning Effort' (the thinking equivalent) where you can set it to Low, Medium, and High, for certain models
There is a bit of nuance to the thinking mode:
• This is different to using the Think Tool • You cannot get access to the extended thinking mode if using OpenRouter chat node • In the Anthropic Dev Docs, they speak about how you can use the extended thinking mode to see how the model processes your prompt (link below) • In the same Docs, Anthropic says if you want to give the model <1k tokens to 'think', you should instead use CoT (chain of thought) prompting using <thinking> XML tags (link below)
I regularly go through the n8n updates page to see what new stuff comes out. You should too!
I also made this video going into a bit more depth about what was mentioned in this post
Bart Slodyczka
Flick through the images and you will see that the Anthropic chat node has a setting where you can 'Enable Thinking'
When you turn this on your agent actually spends tokens 'thinking' about your question, just like what you get in the Claude chat interface.
The last image shows how the output response contains two objects:
• One is the type: thinking - the thinking tokens
• The other is type: text - the actual response
The OpenAI chat node also has a similar setting for 'Reasoning Effort' (the thinking equivalent) where you can set it to Low, Medium, and High, for certain models
There is a bit of nuance to the thinking mode:
• This is different to using the Think Tool
• You cannot get access to the extended thinking mode if using OpenRouter chat node
• In the Anthropic Dev Docs, they speak about how you can use the extended thinking mode to see how the model processes your prompt (link below)
• In the same Docs, Anthropic says if you want to give the model <1k tokens to 'think', you should instead use CoT (chain of thought) prompting using <thinking> XML tags (link below)
I regularly go through the n8n updates page to see what new stuff comes out. You should too!
I also made this video going into a bit more depth about what was mentioned in this post
Video link: https://youtu.be/eEIdHWrsaNc
Anthropic link 1: docs.claude.com/en/docs/build-with-claude/prompt-e…
Anthropic link 2: docs.claude.com/en/docs/build-with-claude/prompt-e…
1 month ago | [YT] | 50